Reflexive verbs in Hebrew: Deep unaccusativity meets lexical semantics

Reflexive verbs in Modern Hebrew show specific morphological marking: only one of the seven verbal templates in the language can be used for reflexives. Yet this morphological marking also appears on anticausative verbs, which have different syntactic and semantic properties. I provide an analysis of reflexivity in Hebrew which does not make reference to dedicated reflexive morphosyntax. By combining independently needed functional heads, the proposal explains what in the syntax underlies this morphology and how different kinds of verbs end up with identical morphophonological properties. To this end, I consider the lexical semantics of individual lexical roots as well as the syntactic configurations in which roots and arguments are embedded. The resulting theory is one in which lexical roots trigger specific interpretations of the syntax at the interfaces.


Background
Reflexive verbs have posed a long-standing puzzle for theories of argument structure: one argument appears to have two thematic roles, agent and patient. If John kicked, John could not have kicked himself, but if John shaved, it is clear that he shaved himself. The reflexive reading for shave arises even without a reflexive anaphor. While some languages, like English, do not differentiate morphologically between verbs like shave and verbs like kick, many languages do express reflexivity through morphological means. The degree to which this state of affairs is problematic varies from theory to theory, but reflexive verbs are predominantly marked morphologically, suggesting that the morphosyntax might be marked as well.
Within contemporary generative work two questions regarding reflexive verbs have risen to the fore. The first is whether there exist dedicated reflexivizers, operators whose sole job is to reduce the arity of a predicate, or whether this job is carried out through a conspiracy of other components of the grammar. Both options carry implications for where the origin of morphological marking lies and what it tracks. If dedicated reflexivizers are part of the morphological toolkit of any grammar, we might expect to find them in many languages. In contrast, if reflexive marking arises through a combination of other means, we would need to identify what these are on a language-by-language basis.
The second question is whether reflexive verbs are unaccusative or unergative: where is the argument generated and how does it come to be the subject of the clause. This question is tied to the environments that are licensed by a reflexive verb, as we will see below.
In general, the answers to both of these questions may well vary by language. On the question of dedicated reflexivizers, Reinhart & Siloni (2005) and Labelle (2008) answer in the affirmative for Hebrew and French, while Lidz (2001) answers in the negative for Kannada. Analyses of reflexive constructions without reflexivizers have been put forward in other languages as well, including Greek (Spathas et al. 2015) and Latin (Miller 2010). The verbal morphology of Modern Hebrew can shed further light on these debates since argument structure alternations are reflected in the templatic morphology of the language.
The current paper presents a novel analysis of reflexives in Hebrew, one that does not make use of a reflexivizer as such and that treats reflexives as unaccusative. Any analysis of Hebrew requires an understanding of how the non-concatenative morphological system is derived. The current account of reflexive verbs is couched in a general theory of the Semitic verb, employing contemporary theories of morphology in order to analyze a peculiarity of Hebrew: reflexive verbs are only possible in one of the verbal templates, specifically the most complex one morphophonologically.
The paper is organized as follows. I first provide a brief introduction to the verbal morphology of Hebrew. Section 2 then surveys and analyzes anticausative verbs in the template in question, hitXaY̯ eZ, and Section 3 does the same for reflexive verbs. Section 4 extends the discussion to the role of deep unaccusativity, surface unaccusativity and lexical semantics in the study of argument structure. Alternative analyses are considered next in Sections 5 (for different theoretical approaches) and 6 (for different approaches to Hebrew), before Section 7 concludes.

Hebrew morphology in a nutshell
Verbs in Modern Hebrew are instantiated in one of seven distinct morphophonological templates. Not all will be discussed here: our focus is mainly on the one notated hitXaY̯ eZ, with other templates brought up as necessary. Each template is traditionally associated with a certain kind of argument structure alternation: causative, passive, and so on ( Berman 1978;Doron 2003;Arad 2005;Borer 2013;Kastner 2016). For example, one typical alternation is between transitive verbs in XaYaZ and their anticausative (detransitivized) variants in niXYaZ. On the notation used here, X, Y and Z are placeholders for the consonants which make up the root,

Reflexive verbs in Hebrew 3
hitXaY̯ eZ, with other templates brought up as necessary. Each template is traditionally associated with a certain kind of argument structure alternation: causative, passive, and so on (Berman 1978;Doron 2003;Arad 2005;Borer 2013;Kastner 2016). For example, one typical alternation is between transitive verbs in XaYaZ and their anticausative (detransitivized) variants in niXYaZ. On the notation used here , X, Y and Z are placeholders for the consonants which make up the root, √ XYZ; examples are given in IPA. The anticausative alternation for XaYaZ∼niXYaZ is shown in (1) 'The gate opened.' Simple as this alternation may be, it does not generalize to all roots. First, there exist anticausative verbs in niXYaZ, (2), with no causative alternation in XaYaZ from which they could have been derived. While causative variants exist in other templates, it is debatable whether a strict derivational relationship should be postulated as in e.g. Laks (2013). Second, there exist verbs in niXYaZ which are not anticausative, (3): they can be shown to pattern with unergative verbs, rather than unaccusative ones (Kastner 2016). In addition, not all roots instantiate verbs in all seven templates, indicating a high degree of lexical idiosyncrasy: for each root, information must be listed indicating which template it can appear in. Simple as this alternation may be, it does not generalize to all roots. First, there exist anticausative verbs in niXYaZ, (2), with no causative alternation in XaYaZ from which they could have been derived. While causative variants exist in other templates, it is debatable whether a strict derivational relationship should be postulated as in e.g. Laks (2013). Second, there exist verbs in niXYaZ which are not anticausative, (3): they can be shown to pattern with unergative verbs, rather than unaccusative ones (Kastner 2016). In addition, not all roots instantiate verbs in all seven templates, indicating a high degree of lexical idiosyncrasy: for each root, information must be listed indicating which template it can appear in.
In addition, not all roots instantiate verbs in all seven templates, indicating a high degree of lexical idiosyncrasy: for each root, information must be listed indicating which template it can appear in.
√ rg' -nirga 'calmed down' (3) gilad Gilad nixnas entered.mid be-gaava in-pride la-bajt to.the-house ha-xadaʃ. the-new 'Gilad entered his new house proudly. ' Of interest in the current paper is the "complex" template hitXaY̯ eZ, exemplified in (4)-(6). The typical alternation for this template is an anticausative one, between a transitive verb in XiY̯ eZ and an anticausative in hitXaY̯ eZ (the notation Y̯ indicates lack of spirantization, a phonological process I return to in Section 3). We will focus on the fact that verbs in hitXaY̯ eZ can have other readings associated with them, (5)-(6), besides an anticausative correspondent of XiY̯ eZ.
(2013). Second, there exist verbs in niXYaZ which are not anticausative, (3): they can be shown to pattern with unergative verbs, rather than unaccusative ones (Kastner 2016). In addition, not all roots instantiate verbs in all seven templates, indicating a high degree of lexical idiosyncrasy: for each root, information must be listed indicating which template it can appear in.
√ rg' -nirga 'calmed down' (3) gilad Gilad nixnas entered.mid be-gaava in-pride la-bajt to.the-house ha-xadaʃ. the-new 'Gilad entered his new house proudly. ' Of interest in the current paper is the "complex" template hitXaY̯ eZ, exemplified in (4)-(6). The typical alternation for this template is an anticausative one, between a transitive verb in XiY̯ eZ and an anticausative in hitXaY̯ eZ (the notation Y̯ indicates lack of spirantization, a phonological process I return to in Section 3). We will focus on the fact that verbs in hitXaY̯ eZ can have other readings associated with them, (5)-(6), besides an anticausative correspondent of XiY̯ eZ.
(2013). Second, there exist verbs in niXYaZ which are not anticausative, (3): they can be shown to pattern with unergative verbs, rather than unaccusative ones (Kastner 2016). In addition, not all roots instantiate verbs in all seven templates, indicating a high degree of lexical idiosyncrasy: for each root, information must be listed indicating which template it can appear in.
√ rg' -nirga 'calmed down' (3) gilad Gilad nixnas entered.mid be-gaava in-pride la-bajt to.the-house ha-xadaʃ. the-new 'Gilad entered his new house proudly. ' Of interest in the current paper is the "complex" template hitXaY̯ eZ, exemplified in (4)-(6). The typical alternation for this template is an anticausative one, between a transitive verb in XiY̯ eZ and an anticausative in hitXaY̯ eZ (the notation Y̯ indicates lack of spirantization, a phonological process I return to in Section 3). We will focus on the fact that verbs in hitXaY̯ eZ can have other readings associated with them, (5)-(6), besides an anticausative correspondent of XiY̯ eZ.
nirga 'calmed down' (3) gilad nixnas be-gaava la-bajt ha-xadaʃ. Gilad entered.mid in-pride to.the-house the-new 'Gilad entered his new house proudly.' Of interest in the current paper is the "complex" template hitXaY̯ eZ, exemplified in (4)-(6). The typical alternation for this template is an anticausative one, between a transitive verb in XiY̯ eZ and an anticausative in hitXaY̯ eZ (the notation Y̯ indicates lack of spirantization, a phonological process I return to in Section 3). We will focus on the fact that verbs in hitXaY̯ eZ can have other readings associated with them, (5)-(6), besides an anticausative correspondent of XiY̯ eZ.
(4) Anticausative a. josi biʃel marak. Yossi cooked.intns soup 'Yossi cooked some soup.' b. ha-marak hitbaʃel ba-ʃemeʃ. the-soup got.cooked.intns.mid in.the-sun 'The soup cooked in the sun.' The puzzle posed by hitXaY̯ eZ is the following: why is it that reflexive verbs appear only in this template and not in any of the others? This question is inherently tied to two related questions: why is this template morphophonologically complex, and what is the range of verbs that may be instantiated in it. To answer these questions, I will propose that reflexives and anticausatives share an unaccusative structure, but that the lexical semantics of the root constrains the derivation. Specifically, reflexive verbs are argued to be the result of unaccusative syntax with an agentive modifier and selforiented lexical semantics. These notions will be made explicit in Sections 2.3 and 3.2. The thrust of the argument is that reflexive readings fall out naturally once certain elements are combined in the syntax, elements which are independently attested in the grammar.

The building blocks of argument structure
The discussion in this paper highlights how roots place requirements on the syntactic derivation. In English, for instance, it has been suggested in different ways that there is a difference between the semantics of √ destroy, √ grow and √ break which goes beyond pure meaning. This difference leads to an inability to take complements in nominalized form (Chomsky 1970;Marantz 1997).
(7) √ destroy: change of state, externally caused a. The enemy's destruction of the city.
b. The city's destruction (by the enemy).
b. *The glass' break. Similar observations have been made more recently for a variety of phenomena in different languages (Haspelmath 1993;Levin & Rappaport Hovav 1995;Schäfer 2008). The details are less important right now than the intuition that something about the lexical semantics of the root constrains what should otherwise be an identical syntactic derivation. In these cases, the underlying assumption is that the morphosyntax of the verbs destroy, grow and break is identical in that they are all made up of a root and a verbalizer, with no extra syntactic material determining their argument structure.
Nevertheless, argument structure alternations can be conditioned by additional syntactic material. For instance, markers such as German sich and Romance se famously reduce the total arity of the verb, descriptively speaking (e.g. Labelle 2008;Schäfer 2008;Cuervo 2014). In order to account for the Hebrew facts, I will take the connection highlighted in the previous section-that between argument structure and the templateand cash it out in terms of the syntactic head Voice.
Throughout the paper I assume that morphological structure is built up in the syntax (Halle & Marantz 1993), with late insertion of phonological material proceeding from the most deeply embedded element outwards (Bobaljik 2000;Embick 2010). The external argument is introduced by the functional head Voice (Kratzer 1996;Pylkkänen 2008). Acategorial roots modify one of the "categorizing" heads v, n and a (Marantz 1997;Arad 2003;Harley 2014). To see how roots affect argument structure, we begin with anticausatives.

The building blocks of argument structure
The discussion in this paper highlights how roots place requirements on the syntactic derivation. In English, for instance, it has been suggested in different ways that there is a difference between the semantics of √ destroy, √ grow and √ break which goes beyond pure meaning. This difference leads to an inability to take complements in nominalized form (Chomsky 1970;Marantz 1997).
(7) √ destroy: change of state, externally caused a. The enemy's destruction of the city.
b. The city's destruction (by the enemy).
b. *The glass' break. Similar observations have been made more recently for a variety of phenomena in different languages (Haspelmath 1993;Levin & Rappaport Hovav 1995;Schäfer 2008). The details are less important right now than the intuition that something about the lexical semantics of the root constrains what should otherwise be an identical syntactic derivation. In these cases, the underlying assumption is that the morphosyntax of the verbs destroy, grow and break is identical in that they are all made up of a root and a verbalizer, with no extra syntactic material determining their argument structure.
Nevertheless, argument structure alternations can be conditioned by additional syntactic material. For instance, markers such as German sich and Romance se famously reduce the total arity of the verb, descriptively speaking (e.g. Labelle 2008;Schäfer 2008;Cuervo 2014). In order to account for the Hebrew facts, I will take the connection highlighted in the previous section-that between argument structure and the templateand cash it out in terms of the syntactic head Voice.
Throughout the paper I assume that morphological structure is built up in the syntax (Halle & Marantz 1993), with late insertion of phonological material proceeding from the most deeply embedded element outwards (Bobaljik 2000;Embick 2010). The external argument is introduced by the functional head Voice (Kratzer 1996;Pylkkänen 2008). Acategorial roots modify one of the "categorizing" heads v, n and a (Marantz 1997;Arad 2003;Harley 2014). To see how roots affect argument structure, we begin with anticausatives. and Reflexive verbs in Hebrew 5

The building blocks of argument structure
The discussion in this paper highlights how roots place requirements on the syntactic derivation. In English, for instance, it has been suggested in different ways that there is a difference between the semantics of √ destroy, √ grow and √ break which goes beyond pure meaning. This difference leads to an inability to take complements in nominalized form (Chomsky 1970;Marantz 1997). (7) √ destroy: change of state, externally caused a. The enemy's destruction of the city.
b. The city's destruction (by the enemy).
b. * The glass' break. Similar observations have been made more recently for a variety of phenomena in different languages (Haspelmath 1993;Levin & Rappaport Hovav 1995;Schäfer 2008). The details are less important right now than the intuition that something about the lexical semantics of the root constrains what should otherwise be an identical syntactic derivation. In these cases, the underlying assumption is that the morphosyntax of the verbs destroy, grow and break is identical in that they are all made up of a root and a verbalizer, with no extra syntactic material determining their argument structure.
Nevertheless, argument structure alternations can be conditioned by additional syntactic material. For instance, markers such as German sich and Romance se famously reduce the total arity of the verb, descriptively speaking (e.g. Labelle 2008;Schäfer 2008;Cuervo 2014). In order to account for the Hebrew facts, I will take the connection highlighted in the previous section-that between argument structure and the templateand cash it out in terms of the syntactic head Voice.
Throughout the paper I assume that morphological structure is built up in the syntax (Halle & Marantz 1993), with late insertion of phonological material proceeding from the most deeply embedded element outwards (Bobaljik 2000;Embick 2010). The external argument is introduced by the functional head Voice (Kratzer 1996;Pylkkänen 2008). Acategorial roots modify one of the "categorizing" heads v, n and a (Marantz 1997;Arad 2003;Harley 2014). To see how roots affect argument structure, we begin with anticausatives.
which goes beyond pure meaning. This difference leads to an inability to take complements in nominalized form (Chomsky 1970;Marantz 1997 Levin & Rappaport Hovav 1995;Schäfer 2008). The portant right now than the intuition that something about the lexical root constrains what should otherwise be an identical syntactic derivases, the underlying assumption is that the morphosyntax of the verbs break is identical in that they are all made up of a root and a verbalizer, tactic material determining their argument structure. , argument structure alternations can be conditioned by additional synor instance, markers such as German sich and Romance se famously reity of the verb, descriptively speaking (e.g. Labelle 2008; Schäfer 2008; order to account for the Hebrew facts, I will take the connection highevious section-that between argument structure and the templateterms of the syntactic head Voice. the paper I assume that morphological structure is built up in the syntax z 1993), with late insertion of phonological material proceeding from embedded element outwards (Bobaljik 2000;Embick 2010). The exis introduced by the functional head Voice (Kratzer 1996; Pylkkänen ial roots modify one of the "categorizing" heads v, n and a Harley 2014). To see how roots affect argument structure, we begin es.
: change of state, externally caused a. The enemy's destruction of the city. b. The city's destruction (by the enemy). . en made more recently for a variety of phenomena in differ-993; Levin & Rappaport Hovav 1995;Schäfer 2008). The ht now than the intuition that something about the lexical ins what should otherwise be an identical syntactic derivaerlying assumption is that the morphosyntax of the verbs ntical in that they are all made up of a root and a verbalizer, rial determining their argument structure. tructure alternations can be conditioned by additional synmarkers such as German sich and Romance se famously rerb, descriptively speaking (e.g. Labelle 2008; Schäfer 2008; count for the Hebrew facts, I will take the connection highn-that between argument structure and the templatee syntactic head Voice. ssume that morphological structure is built up in the syntax h late insertion of phonological material proceeding from element outwards (Bobaljik 2000;Embick 2010). The exd by the functional head Voice (Kratzer 1996; Pylkkänen dify one of the "categorizing" heads v, n and a (Marantz 4). To see how roots affect argument structure, we begin  (Haspelmath 1993;Levin & Rappaport Hovav 1995;Schäfer 2008). The details are less important right now than the intuition that something about the lexical semantics of the root constrains what should otherwise be an identical syntactic derivation. In these cases, the underlying assumption is that the morphosyntax of the verbs destroy, grow and break is identical in that they are all made up of a root and a verbalizer, with no extra syntactic material determining their argument structure.
Nevertheless, argument structure alternations can be conditioned by additional syntactic material. For instance, markers such as German sich and Romance se famously reduce the total arity of the verb, descriptively speaking (e.g. Labelle 2008;Schäfer 2008;Cuervo 2014). In order to account for the Hebrew facts, I will take the connection highlighted in the previous section-that between argument structure and the template-and cash it out in terms of the syntactic head Voice.
Throughout the paper I assume that morphological structure is built up in the syntax (Halle & Marantz 1993), with late insertion of phonological material proceeding from the most deeply embedded element outwards (Bobaljik 2000;Embick 2010). The external argument is introduced by the functional head Voice (Kratzer 1996;Pylkkänen 2008). Acategorial roots modify one of the "categorizing" heads v, n and a (Marantz 1997;Arad 2003;Harley 2014). To see how roots affect argument structure, we begin with anticausatives.

Anticausatives in Hebrew
The traditional view of Semitic templates is that they encode argument structure alternations, as in (1) and (4)

Anticausatives in Hebrew
The traditional view of Semitic templates is that they encode argument structure alternations, as in (1) and (4)  f. √ bʃl biʃel 'cooked' hitbaʃel 'got cooked' In all these cases, the non-active version is a detransitivized form of the active version and shares the same root as the active verb. The derived verbs in (10) are all intransitive and their bases transitive.

Agent-oriented adverbs
A number of agent-oriented tests show that no agents are possible with anticausatives in Hebrew. The phrase me-aʦmo 'by itself' diagnoses the lack of an external argument (Levin & Rappaport Hovav 1995;Alexiadou & Anagnostopoulou 2004;Alexiadou & Doron 2012), be it explicit (as with transitive verbs) or implicit (as with passives). The test is appropriate with anticausatives in the niXYaZ template shown above, (11a), but not with direct objects of transitive verbs, (11b), or with passive verbs, (11c-d

Anticausatives in Hebrew
The traditional view of Semitic templates is that they encode argument structure alternations, as in (1) and (4)  f. √ bʃl biʃel 'cooked' hitbaʃel 'got cooked' In all these cases, the non-active version is a detransitivized form of the active version and shares the same root as the active verb. The derived verbs in (10) are all intransitive and their bases transitive.

Agent-oriented adverbs
A number of agent-oriented tests show that no agents are possible with anticausatives in Hebrew. The phrase me-aʦmo 'by itself' diagnoses the lack of an external argument (Levin & Rappaport Hovav 1995;Alexiadou & Anagnostopoulou 2004;Alexiadou & Doron 2012), be it explicit (as with transitive verbs) or implicit (as with passives). The test is appropriate with anticausatives in the niXYaZ template shown above, (11a), but not with direct objects of transitive verbs, (11b), or with passive verbs, (11c-d

Anticausatives in Hebrew
The traditional view of Semitic templates is that they encode argument structure alternations, as in (1) and (4)  f. √ bʃl biʃel 'cooked' hitbaʃel 'got cooked' In all these cases, the non-active version is a detransitivized form of the active version and shares the same root as the active verb. The derived verbs in (10) are all intransitive and their bases transitive.

Agent-oriented adverbs
A number of agent-oriented tests show that no agents are possible with anticausatives in Hebrew. The phrase me-aʦmo 'by itself' diagnoses the lack of an external argument (Levin & Rappaport Hovav 1995;Alexiadou & Anagnostopoulou 2004;Alexiadou & Doron 2012), be it explicit (as with transitive verbs) or implicit (as with passives). The test is appropriate with anticausatives in the niXYaZ template shown above, (11a), but not with direct objects of transitive verbs, (11b), or with passive verbs, (11c-d

Anticausatives in Hebrew
The traditional view of Semitic templates is that they encode argument structure alternations, as in (1) and (4)  f. √ bʃl biʃel 'cooked' hitbaʃel 'got cooked' In all these cases, the non-active version is a detransitivized form of the active version and shares the same root as the active verb. The derived verbs in (10) are all intransitive and their bases transitive.

Agent-oriented adverbs
A number of agent-oriented tests show that no agents are possible with anticausatives in Hebrew. The phrase me-aʦmo 'by itself' diagnoses the lack of an external argument (Levin & Rappaport Hovav 1995;Alexiadou & Anagnostopoulou 2004;Alexiadou & Doron 2012), be it explicit (as with transitive verbs) or implicit (as with passives). The test is appropriate with anticausatives in the niXYaZ template shown above, (11a), but not with direct objects of transitive verbs, (11b), or with passive verbs, (11c-d

Anticausatives in Hebrew
The traditional view of Semitic templates is that they encode argument structure alternations, as in (1) and (4)  √ bʃl biʃel 'cooked' hitbaʃel 'got cooked' In all these cases, the non-active version is a detransitivized form of the active version and shares the same root as the active verb. The derived verbs in (10) are all intransitive and their bases transitive.

Agent-oriented adverbs
A number of agent-oriented tests show that no agents are possible with anticausatives in Hebrew. The phrase me-aʦmo 'by itself' diagnoses the lack of an external argument (Levin & Rappaport Hovav 1995;Alexiadou & Anagnostopoulou 2004;Alexiadou & Doron 2012), be it explicit (as with transitive verbs) or implicit (as with passives). The test is appropriate with anticausatives in the niXYaZ template shown above, (11a), but not with direct objects of transitive verbs, (11b), or with passive verbs, (11c-d

Anticausatives in Hebrew
The traditional view of Semitic templates is that they encode argument structure alternations, as in (1) and (4)  f. √ bʃl biʃel 'cooked' hitbaʃel 'got cooked' In all these cases, the non-active version is a detransitivized form of the active version and shares the same root as the active verb. The derived verbs in (10) are all intransitive and their bases transitive.

Agent-oriented adverbs
A number of agent-oriented tests show that no agents are possible with anticausatives in Hebrew. The phrase me-aʦmo 'by itself' diagnoses the lack of an external argument (Levin & Rappaport Hovav 1995;Alexiadou & Anagnostopoulou 2004;Alexiadou & Doron 2012), be it explicit (as with transitive verbs) or implicit (as with passives). The test is appropriate with anticausatives in the niXYaZ template shown above, (11a), but not with direct objects of transitive verbs, (11b), or with passive verbs, (11c-d In all these cases, the non-active version is a detransitivized form of the active version and shares the same root as the active verb. The derived verbs in (10) are all intransitive and their bases transitive.

Agent-oriented adverbs
A number of agent-oriented tests show that no agents are possible with anticausatives in Hebrew. The phrase me-aʦmo 'by itself' diagnoses the lack of an external argument (Levin & Rappaport Hovav 1995;Alexiadou & Anagnostopoulou 2004;Alexiadou & Doron 2012), be it explicit (as with transitive verbs) or implicit (as with passives). The test is appropriate with anticausatives in the niXYaZ template shown above, (11a), but not with direct objects of transitive verbs, (11b), or with passive verbs, (11c-d).
(12) a. ha-ʦoref pirek et ha-ʦamid. the-jeweler dismantled.intns acc the-bracelet 'The jeweler took the bracelet apart.' b. ha-ʦamid hitparek me-aʦmo. the-bracelet dismantled.intns.mid from-itself 'The bracelet fell apart of its own accord.' Other traditional tests such as incompatibility with by-phrases and agent-oriented adverbs support the claim that the derived verbs are indeed unaccusative, (13). 2 (13) *ha-ʦamid hitparek { al-jedej ha-tsoref / be-mejomanut }. the-bracelet fell.apart.intns.mid by the-jeweler in-skill (int. 'The bracelet was dismantled by the jeweler/skillfully') Two language-specific diagnostics have also been proposed in the literature: VS order and the possessive dative. These tests align with the ones above, as shown next.

Order of subject and verb
The word order of Modern Hebrew is typically SV(O) as seen in all of the examples above, but unaccusatives allow the verb to appear before the underlying object, (14a). Presumably this is because the underlying object stays low in its base-generated position, (14b). Unergatives do not allow VS order except for the marked structure known as "stylistic inversion", (15); see Shlonsky (1987), Shlonsky & Doron (1991) and Borer (1995). (14) a. nafl-u ʃaloʃ kosot be-ʃmone ba-boker. ✓ internal argument fell.smpl-3pl three glasses in-eight in.the-morning 'Three glasses fell at 8am.' b. b.
ha-ʦamid the-bracelet hitparek dismantled.intns.mid me-aʦmo. from-itself 'The bracelet fell apart of its own accord.' Other traditional tests such as incompatibility with by-phrases and agent-oriented adverbs support the claim that the derived verbs are indeed unaccusative, (13 (int. 'The bracelet was dismantled by the jeweler/skillfully') Two language-specific diagnostics have also been proposed in the literature: VS order and the possessive dative. These tests align with the ones above, as shown next.

Order of subject and verb
The word order of Modern Hebrew is typically SV(O) as seen in all of the examples above, but unaccusatives allow the verb to appear before the underlying object, (14a). Presumably this is because the underlying object stays low in its base-generated position, (14b). Unergatives do not allow VS order except for the marked structure known as "stylistic inversion", (15); see Shlonsky (1987), Shlonsky & Doron (1991) and Borer (1995). (15) #navx-u ʃloʃa klavim be-ʃmone ba-boker. ✗ external argument barked.smpl-3pl three dogs in-eight in.the-morning 'And thence barked three dogs at 8am.' (Marked variant) Anticausatives in hitXaY̯ eZ allow VS order just like their counterparts in niXYaZ: (16) patterns with (14).

Possessive datives
The second diagnostic is the possessive dative, a type of possessor raising. This construction has been claimed to only be possible with internal arguments (Borer & Grodzinsky 1986), though we return to critiques of it in Section 4.1.2.
A simple unaccusative like nafal 'fell' in the underspecified XaYaZ template is compatible with a possessive dative, (17a), as is a transitive construction, (17b), whereas an unergative verb leads to a deviant, affected interpretation, (17c). Anticausatives in hitXaY̯ eZ are compatible with possessive datives, (18). hitparek l-i ha-ʃaon. ✓ internal argument dismantled.intns.mid to-me the-watch 'My watch broke.' Having confirmed that anticausatives in hitXaY̯ eZ are indeed unaccusative, I turn to the formalization.

Analysis
As noted earlier, there are seven distinct morphophonological verbal forms in Modern Hebrew, the "templates". A few words on the system as a whole are in order before focusing on the analysis of an individual template. It has been suggested that analysis of the templates can rely on different values of the Voice head (Doron 2003;Arad 2005;Kastner 2016), originally proposed as the functional head introducing external arguments (Kratzer 1996). The typology of Voice heads follows recent proposals by Schäfer (2008) for German, Bruening (2014) for English, Alexiadou & Doron (2012) and Spathas et al. (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, and Kastner & Zu (2015) for Latin. What this kind of framework does is to separate the syntactic and semantic behavior of Voice heads, leading to new analyses of argument structure alternations Schäfer 2017).
I take Voice to be instantiated in three different ways, depending on whether it has a [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, either requiring a DP in the specifier of its head, prohibiting a DP in its specifier or remaining agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the structure without a canonical subject.

(19) a. Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991).
b. ⟦Voice [-D] ⟧= λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all languages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016(Kastner , 2017 Adding an agentive modifier, guages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016;. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two additional templates which we have already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ and hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I explain how it brings about an external argument without requiring one in the syntax (which is what the feature [+D] does, but Voice [+D] will not be discussed further.).
3 Notational matters: the template heXYiZ usually appears in the literature as hiXYiZ, with the first vowel an /i/. Speakers of my generation and at least one generation older use the /e/ form, and so I transcribe /e/ throughout. Conversely, the initial /h/ is usually dropped in speech. I retain it for two reasons. First, /h/ is still pronounced by some older speakers and certain sociolinguistic groups, especially marginalized ones (Schwarzwald 1981;Gafter 2014b , derives the two additional templates which we have already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ and hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I explain how it brings about an external argument without requiring one in the syntax (which is what the feature [+D] does, but Voice [+D] will not be discussed further.).
Combined with the two additional passive templates-which are derived by use of an additional Pass head (Alexiadou & Doron 2012)-we can generate all seven templates of Hebrew in the syntax.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991 Voice [-D] A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991 Putting aside the exact morphophonological processes, here is how these functional heads work in the syntax. Anticausative verbs are derived by taking an existing transitive vP (one that has a direct object) and merging Voice [-D] , thereby detransitivizing the verb. This results in anticausative alternations as in (23a-b), between 6

Anticausatives in Hebrew
The traditional view of Semitic templates is that they encode ar nations, as in (1) and (4)  f. √ bʃl biʃel 'cooked' In all these cases, the non-active version is a detransitivized sion and shares the same root as the active verb. The derived ver sitive and their bases transitive. with Voice and with Voice [-D] , and in (24a-b), between 6 Kastner

Anticausatives in Hebrew
The traditional view of Semitic templates is that they encode argument structure alternations, as in (1) and (4)  f. √ bʃl biʃel 'cooked' hitbaʃel 'got cooked' In all these cases, the non-active version is a detransitivized form of the active version and shares the same root as the active verb. The derived verbs in (10) are all intransitive and their bases transitive.

Agent-oriented adverbs
A number of agent-oriented tests show that no agents are possible with anticausatives in Hebrew. The phrase me-aʦmo 'by itself' diagnoses the lack of an external argument (Levin & Rappaport Hovav 1995;Alexiadou & Anagnostopoulou 2004;Alexiadou & Doron 2012), be it explicit (as with transitive verbs) or implicit (as with passives). The test is appropriate with anticausatives in the niXYaZ template shown above, (11a), but not with direct objects of transitive verbs, (11b), or with passive verbs, (11c-d). with Voice + ner 2016), originally proposed as the functional head introducing exter (Kratzer 1996). The typology of Voice heads follows recent proposals by for German, Bruening (2014) for English, Alexiadou & Doron (2012 for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015 andZu (2015) for Latin. What this kind of framework does is syntactic and semantic behavior of Voice heads, leading to new analyse structure alternations Schäfer 2017).
I take Voice to be instantiated in three different ways, depending on w [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP requiring a DP in the specifier of its head, prohibiting a DP in its spec ing agnostic about the specifier. In the discussion of hitXaY̯ eZ we will Voice [-D] , a Voice head which bans the merger of a DP in its specifier, lea ture without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything wit from merging in its specifier. As typically assumed for unaccusative little v or unaccusative does not assign accusative case itself (Chomsky 1995) or t culation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instant guages. In Hebrew, it has been argued that the three variants have dist (Kastner 2016;. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two additional te we have already seen in (4)-(6) and which lie at the heart of the current st hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where it brings about an external argument without requiring one in the syntax the feature [+D] does, but Voice [+D] will not be discussed further.).
3 Notational matters: the template heXYiZ usually appears in the literature as hiXYiZ, w an /i/. Speakers of my generation and at least one generation older use the /e/ form, and throughout. Conversely, the initial /h/ is usually dropped in speech. I retain it for two re still pronounced by some older speakers and certain sociolinguistic groups, especially m (Schwarzwald 1981;Gafter 2014b 2016), originally proposed as the functional head introducing external arguments tzer 1996). The typology of Voice heads follows recent proposals by Schäfer (2008) erman, Bruening (2014) for English, Alexiadou & Doron (2012) and Spathas et al. 5) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, Kastner & Zu (2015) for Latin. What this kind of framework does is to separate the actic and semantic behavior of Voice heads, leading to new analyses of argument cture alternations Schäfer 2017). I take Voice to be instantiated in three different ways, depending on whether it has a feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, either iring a DP in the specifier of its head, prohibiting a DP in its specifier or remaingnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to e [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the strucwithout a canonical subject. a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all lanes. In Hebrew, it has been argued that the three variants have distinct exponents tner 2016; 2017). a.
Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two additional templates which ave already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ and aY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I explain how ings about an external argument without requiring one in the syntax (which is what eature [+D] does, but Voice [+D] will not be discussed further.).
tional matters: the template heXYiZ usually appears in the literature as hiXYiZ, with the first vowel . Speakers of my generation and at least one generation older use the /e/ form, and so I transcribe /e/ ghout. Conversely, the initial /h/ is usually dropped in speech. I retain it for two reasons. First, /h/ is ronounced by some older speakers and certain sociolinguistic groups, especially marginalized ones arzwald 1981; Gafter 2014b). Second, the initial h-should help the non-Semitist reader distinguish emplate from others. ears that √ action does not combine with Voice [+D] ; see Kastner (2016: Chapter 2.4.1) for discus-. No external argument can be merged in the specifier of Voice [-D] , rendering the structures in (23b) and (24b) unaccusative.

Kastner
Combined with the two additional passive templates-which are derived by use of an additional Pass head (Alexiadou & Doron 2012)-we can generate all seven templates of Hebrew in the syntax. (21) a. [ Focusing back on hitXaY̯ eZ, it is evident that hit-is a prefix, rather than a higher dummy DP or clitic (like French se in Labelle 2008 or German sich in Schäfer 2008) since its form is conditioned by tense and agreement, (22), a hallmark of agreement affixes (Nevins 2011).
Putting aside the exact morphophonological processes, here is how these functional heads work in the syntax. Anticausative verbs are derived by taking an existing transitive vP (one that has a direct object) and merging Voice [-D] , thereby detransitivizing the verb. This results in anticausative alternations as in (23a-b), between √ ʃbr with Voice and with Voice [-D] , and in (24a-b), between √ prk with Voice+ √ action and with Combined with the two additional passive templates-which are derived by use of an additional Pass head (Alexiadou & Doron 2012)-we can generate all seven templates of Hebrew in the syntax. (21) a. [ Focusing back on hitXaY̯ eZ, it is evident that hit-is a prefix, rather than a higher dummy DP or clitic (like French se in Labelle 2008 or German sich in Schäfer 2008) since its form is conditioned by tense and agreement, (22), a hallmark of agreement affixes (Nevins 2011).
Putting aside the exact morphophonological processes, here is how these functional heads work in the syntax. Anticausative verbs are derived by taking an existing transitive vP (one that has a direct object) and merging Voice [-D] , thereby detransitivizing the verb. This results in anticausative alternations as in (23a-b), between √ ʃbr with Voice and with Voice [-D] , and in (24a-b), between √ prk with Voice+ √ action and with ner 2016), originally proposed as the functional head introducing external arguments (Kratzer 1996). The typology of Voice heads follows recent proposals by Schäfer (2008) for German, Bruening (2014) for English, Alexiadou & Doron (2012) and Spathas et al. (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, and Kastner & Zu (2015) for Latin. What this kind of framework does is to separate the syntactic and semantic behavior of Voice heads, leading to new analyses of argument structure alternations Schäfer 2017). I take Voice to be instantiated in three different ways, depending on whether it has a [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, either requiring a DP in the specifier of its head, prohibiting a DP in its specifier or remaining agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the structure without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all languages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016;. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two additional templates which we have already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ and hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I explain how it brings about an external argument without requiring one in the syntax (which is what the feature [+D] does, but Voice [+D] will not be discussed further.).
In (23), Voice introduces an external argument and Voice [-D] blocks one. The same holds for (24) where the structure also contains the modifier √ action, whose exact workings will wait for the next section.
In this section I have set up the basic machinery needed to derive argument structure alternations in the language, allowing for a straightforward description of anticausatives using the functional head Voice [-D] . The next section develops the system in order to account for the main empirical puzzle: reflexive verbs in Hebrew appear in only one of the templates, arguably the most marked one.

Reflexives
In what follows I turn to reflexives in hitXaY̯ eZ. It has recently been proposed that dedicated reflexivizers are not necessary in order to derive reflexives in certain languages. I take this claim one step further based on Hebrew, arguing that dedicated reflexivizers are not necessary and that the same functional heads can be used to derive both reflexives and anticausatives, at least in this language. I develop the empirical picture in Section 3.1, present my analysis in Section 3.2 and return to tie a loose anticausative end in Section 3.3.

b. hitXaY̯ eZ, hitparek 'fell apart'
Reflexive verbs in Hebrew 11 In (23), Voice introduces an external argument and Voice [-D] blocks one. The same holds for (24) where the structure also contains the modifier √ action, whose exact workings will wait for the next section.
In this section I have set up the basic machinery needed to derive argument structure alternations in the language, allowing for a straightforward description of anticausatives using the functional head Voice [-D] . The next section develops the system in order to account for the main empirical puzzle: reflexive verbs in Hebrew appear in only one of the templates, arguably the most marked one.

Reflexives
In what follows I turn to reflexives in hitXaY̯ eZ. It has recently been proposed that dedicated reflexivizers are not necessary in order to derive reflexives in certain languages. I take this claim one step further based on Hebrew, arguing that dedicated reflexivizers are not necessary and that the same functional heads can be used to derive both reflexives and anticausatives, at least in this language. I develop the empirical picture in Section 3.1, present my analysis in Section 3.2 and return to tie a loose anticausative end in Section 3.3.
In (23), Voice introduces an external argument and Voice [-D] blocks one. The same holds for (24) where the structure also contains the modifier ner 2016), originally proposed as the functional head introducing external argumen (Kratzer 1996). The typology of Voice heads follows recent proposals by Schäfer (2008for German, Bruening (2014 for English, Alexiadou & Doron (2012) and Spathas et a (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandi and Kastner & Zu (2015) for Latin. What this kind of framework does is to separate th syntactic and semantic behavior of Voice heads, leading to new analyses of argumen structure alternations Schäfer 2017).
I take Voice to be instantiated in three different ways, depending on whether it has [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, eithe requiring a DP in the specifier of its head, prohibiting a DP in its specifier or remai ing agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer t Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the stru ture without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] featur from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [ does not assign accusative case itself (Chomsky 1995) or through the ca culation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all la guages. In Hebrew, it has been argued that the three variants have distinct exponen (Kastner 2016;. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two additional templates whic we have already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ an hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I explain ho it brings about an external argument without requiring one in the syntax (which is wha the feature [+D] does, but Voice [+D] will not be discussed further.).
, whose exact workings will wait for the next section.
In this section I have set up the basic machinery needed to derive argument structure alternations in the language, allowing for a straightforward description of anticausatives using the functional head Voice [-D] . The next section develops the system in order to account for the main empirical puzzle: reflexive verbs in Hebrew appear in only one of the templates, arguably the most marked one.

Reflexives
In what follows I turn to reflexives in hitXaY̯ eZ. It has recently been proposed that dedicated reflexivizers are not necessary in order to derive reflexives in certain languages. I take this claim one step further based on Hebrew, arguing that dedicated reflexivizers are not necessary and that the same functional heads can be used to derive both reflexives and anticausatives, at least in this language. I develop the empirical picture in Section 3.1, present my analysis in Section 3.2 and return to tie a loose anticausative end in Section 3.3.

Background
The main phenomenon analyzed in this paper is as follows. The verbal template hitXaY̯ eZ shows the same morphological marking for reflexives and anticausatives. By "reflexive verb" in this article I mean (25): Canonical reflexive verb (i) A monovalent verb whose DP internal argument X is interpreted as both Agent and Theme, and (ii) where no other argument Y (implicit or explicit) can be interpreted as Agent or Theme, and (iii) where the structure involves no pronominal elements such as himself.
With this definition we focus on reflexives that are morphologically marked, rather than other reflexive strategies such as anaphora. These verbs in Hebrew are only attested in hitXaY̯ eZ. A sample is given in (26). 5 This kind of morphology is reminiscent of markers such as Romance se, German sich and Russian -sja. Yet unlike languages like French where se might be ambiguous between a number of readings (reflexive, reciprocal and anticausative), hitXaY̯ eZ is never ambiguous in Hebrew for a given root (this generalization will be qualified in Section 4.2). French se can be used in reflexive, reciprocal and non-active contexts with a variety of predicates:  (Labelle 2008: 835) Le vase se brise. the vase se breaks-3s 'The vase is breaking.' But Hebrew hitXaY̯ eZ is unambiguous. The verb hitlabeʃ 'got dressed' is deterministically reflexive and cannot be used in an anticausative (or reciprocal) context, as shown by its incompatibility with 'by itself' in (28a). In contrast, the verb hitaʦben 'got annoyed' is uniformly anticausative and cannot be used with an agent-oriented adverb such as 'on purpose' (Alexiadou & Anagnostopoulou 2004) in (28b).
(*me-aʦmam) Luc and-Pierre dressed.up.intns.mid-3pl from-themselves 'Luc and Pierre got dressed' [reflexive only] b. ha-saxkan hitaʦben (*be-xavana) kʃe-lo masru lo. the-player got.annoyed.intns.mid on-purpose when-neg passed to.him 'The player got annoyed when he wasn't passed the ball.' I argue below that this contrast ultimately derives from the lexical semantics of the root. Dressing up is usually something one does on oneself, while annoying is usually something that one does to someone else. This notion will be made precise in Section 3.3. For now, note that the root itself is not enough to force a reflexive reading. The root

Reflexive verbs in Hebrew
But Hebrew hitXaY̯ eZ is unambiguo tically reflexive and cannot be used in an by its incompatibility with 'by itself' in (2 is uniformly anticausative and cannot be purpose' ( The point is once again that it is not enough for the root to be compatible with a reflexive reading in order for the verb to be reflexive. In English, for instance, wash and shave do not require any special morphological marking. But in Hebrew, both the root and the template combine to decide the meaning and argument structure of the verb, as I explain next.

Analysis
The intuition behind the analysis is as follows: reflexive verbs in hitXaY̯ eZ consist of an unaccusative structure with extra agentive semantics. This combination is only possible if the internal argument is allowed to saturate the semantic function of an external argument, in a way I formalize below.
My proposed analysis consists of three parts, all independently necessary. The first piece of the puzzle is the non-active Voice head introduced in Section 2.3, Voice [-D] . There, we noted that this head underlies argument structure alternations in a number of languages, including across four different templates in Hebrew (XaYaZ ∼ niXYaZ and XiY̯ eZ ∼ hitXaY̯ eZ). The second piece of the reflexive puzzle is the agentive modifier Hebrew 9 inally proposed as the functional head introducing external arguments The typology of Voice heads follows recent proposals by Schäfer (2008) uening (2014) for English, Alexiadou & Doron (2012) and Spathas et al. k, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, Zu (2015) for Latin. What this kind of framework does is to separate the emantic behavior of Voice heads, leading to new analyses of argument ations Schäfer 2017). to be instantiated in three different ways, depending on whether it has a [-D] feature or is underspecified for [±D]. This is an EPP feature, either in the specifier of its head, prohibiting a DP in its specifier or remainout the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to e head which bans the merger of a DP in its specifier, leaving the strucanonical subject.

ice [-D]
Voice head with a [-D] feature, prohibiting anything with a [D] feature m merging in its specifier. typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] es not assign accusative case itself (Chomsky 1995) or through the callation of dependent case (Marantz 1991).
oice [-D] = λP <s,t> .P ite classification of Voice heads is not necessarily instantiated in all lanrew, it has been argued that the three variants have distinct exponents 2017). ice ↔ XaYaZ ice [+D] ↔ heXYiZ 3 gentive modifier, √ action, derives the two additional templates which seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ and element is discussed in more detail in Section 3.2, where I explain how an external argument without requiring one in the syntax (which is what ] does, but Voice [+D] will not be discussed further.).
s: the template heXYiZ usually appears in the literature as hiXYiZ, with the first vowel my generation and at least one generation older use the /e/ form, and so I transcribe /e/ ersely, the initial /h/ is usually dropped in speech. I retain it for two reasons. First, /h/ is y some older speakers and certain sociolinguistic groups, especially marginalized ones 1; Gafter 2014b). Second, the initial h-should help the non-Semitist reader distinguish others. ction does not combine with Voice [+D] ; see Kastner (2016: Chapter 2.4.1) for discus-, also introduced in Section 2.3 but not elaborated on yet. The third piece is a compositional mechanism operating in the syntax-semantics interface developed by Wood (2014), which I adopt. All three pieces can be shown to be independently needed (not only in Hebrew but for the theory as a whole) and their combination correctly predicts both the syntactic-semantic behavior of hitXaY̯ eZ and its morphophonological makeup.

√action
For this part of the puzzle I will modify the suggestion made by Doron (2003) according to which Hebrew has an agentive modifier with predictable spell-out and consistent semantics. The current section formalizes this element as Reflexive verbs in Hebrew 9 ner 2016), originally proposed as the functional head introducing external arguments (Kratzer 1996). The typology of Voice heads follows recent proposals by Schäfer (2008) for German, Bruening (2014) for English, Alexiadou & Doron (2012) and Spathas et al. (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, and Kastner & Zu (2015) for Latin. What this kind of framework does is to separate the syntactic and semantic behavior of Voice heads, leading to new analyses of argument structure alternations Schäfer 2017). I take Voice to be instantiated in three different ways, depending on whether it has a [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, either requiring a DP in the specifier of its head, prohibiting a DP in its specifier or remaining agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the structure without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all languages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016;. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two additional templates which we have already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ and hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I explain how it brings about an external argument without requiring one in the syntax (which is what the feature [+D] does, but Voice [+D] will not be discussed further.).
and draws comparisons with other languages.
In Hebrew the modifier Reflexive verbs in Hebrew 9 ner 2016), originally proposed as the functional head introducing external arguments (Kratzer 1996). The typology of Voice heads follows recent proposals by Schäfer (2008) for German, Bruening (2014) for English, Alexiadou & Doron (2012) and Spathas et al. (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, and Kastner & Zu (2015) for Latin. What this kind of framework does is to separate the syntactic and semantic behavior of Voice heads, leading to new analyses of argument structure alternations Schäfer 2017). I take Voice to be instantiated in three different ways, depending on whether it has a [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, either requiring a DP in the specifier of its head, prohibiting a DP in its specifier or remaining agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the structure without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all languages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016;. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two additional templates which we have already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ and hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I explain how it brings about an external argument without requiring one in the syntax (which is what the feature [+D] does, but Voice [+D] will not be discussed further.).
3 Notational matters: the template heXYiZ usually appears in the literature as hiXYiZ, with the first vowel an /i/. Speakers of my generation and at least one generation older use the /e/ form, and so I transcribe /e/ throughout. Conversely, the initial /h/ is usually dropped in speech. I retain it for two reasons. First, /h/ is still pronounced by some older speakers and certain sociolinguistic groups, especially marginalized ones (Schwarzwald 1981;Gafter 2014b is attested in hitXaY̯ eZ and in the "intensive" template XiY̯ eZ. Consider first the typical difference between verbs in XaYaZ (with unmarked Voice) and XiY̯ eZ (with Voice and Reflexive verbs in Hebrew 9 ner 2016), originally proposed as the functional head introducing external arguments (Kratzer 1996). The typology of Voice heads follows recent proposals by Schäfer (2008) for German, Bruening (2014) for English, Alexiadou & Doron (2012) and Spathas et al. (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, and Kastner & Zu (2015) for Latin. What this kind of framework does is to separate the syntactic and semantic behavior of Voice heads, leading to new analyses of argument structure alternations Schäfer 2017). I take Voice to be instantiated in three different ways, depending on whether it has a [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, either requiring a DP in the specifier of its head, prohibiting a DP in its specifier or remaining agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the structure without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all languages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016;. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two additional templates which we have already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ and hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I explain how it brings about an external argument without requiring one in the syntax (which is what the feature [+D] does, but Voice [+D] will not be discussed further.).
3 Notational matters: the template heXYiZ usually appears in the literature as hiXYiZ, with the first vowel an /i/. Speakers of my generation and at least one generation older use the /e/ form, and so I transcribe /e/ throughout. Conversely, the initial /h/ is usually dropped in speech. I retain it for two reasons. First, /h/ is still pronounced by some older speakers and certain sociolinguistic groups, especially marginalized ones (Schwarzwald 1981;Gafter 2014b ). In (30a) both agents and causes are possible with the "simple" XaYaZ verb ʃavru 'broke', but in (30b) only the agent is available with the "intensive" XiY̯ eZ verb ʃibru 'broke to bits'. This element is phonologically overt. I follow Doron (2003) and Kastner (2017) in assuming that XaYaZ is derived morphophonologically through the combination of Voice, v and the root, whereas XiY̯ eZ is the result of adding The tripartite classification of Voice heads is not necessarily instantiated in all languages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016;. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two additional templates which we have already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ and hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I explain how it brings about an external argument without requiring one in the syntax (which is what the feature [+D] does, but Voice [+D] will not be discussed further.).
3 Notational matters: the template heXYiZ usually appears in the literature as hiXYiZ, with the first vowel an /i/. Speakers of my generation and at least one generation older use the /e/ form, and so I transcribe /e/ throughout. Conversely, the initial /h/ is usually dropped in speech. I retain it for two reasons. First, /h/ is still pronounced by some older speakers and certain sociolinguistic groups, especially marginalized ones (Schwarzwald 1981;Gafter 2014b Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessar guages. In Hebrew, it has been argued that the three variants (Kastner 2016;. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two ad we have already seen in (4)-(6) and which lie at the heart of the hitXaY̯ eZ. 4 This element is discussed in more detail in Section it brings about an external argument without requiring one in the feature [+D] does, but Voice [+D] will not be discussed furth 3 Notational matters: the template heXYiZ usually appears in the literature an /i/. Speakers of my generation and at least one generation older use the / throughout. Conversely, the initial /h/ is usually dropped in speech. I retain still pronounced by some older speakers and certain sociolinguistic groups (Schwarzwald 1981;Gafter 2014b). Second, the initial h-should help the n this template from others. 4 It appears that √ action does not combine with Voice [+D] ; see Kastner (2 sion. is in the structure. This element has the phonological property that spirantization of the middle root consonant is blocked. In (30a), the medial /b/ of 6 K

Anticausatives in Hebrew
The traditional view of Semitic templates is that they encode argument structure nations, as in (1) and (4)  XaYaZ ∼ niXYaZ hitbaʃel 'g In all these cases, the non-active version is a detransitivized form of the acti sion and shares the same root as the active verb. The derived verbs in (10) are all i sitive and their bases transitive.

Agent-oriented adverbs
A number of agent-oriented tests show that no agents are possible with anticaus in Hebrew. The phrase me-aʦmo 'by itself' diagnoses the lack of an external argu (Levin & Rappaport Hovav 1995;Alexiadou & Anagnostopoulou 2004;Alexiad Doron 2012), be it explicit (as with transitive verbs) or implicit (as with passives test is appropriate with anticausatives in the niXYaZ template shown above, (11a not with direct objects of transitive verbs, (11b), or with passive verbs, (11c-d). An anonymous reviewer asserts that this morphophonological process has no bearing on the internal structure of these verbs. Two considerations lead me to disagree: whether the spirantization process is productive, and whether the blocking is grammatical. On both counts, the answer is affirmative. Temkin Martínez & Ivana Müllner (2016) conducted a nonce word study for Hebrew and found that native speakers normally spirantize the three stops, but do not spirantize them in medial position of XiY̯ eZ, as would be expected. The results were not categorical, however, in line with previous work; Adam (2002) previously identified patterns of variation in the application of the phonological rule and the morphophonological one. Clearly, then, there is a variable phonological process which is blocked by grammatical means, indicating that these grammatical means should be part of the analysis.
As far as the semantics is concerned, the difference in possible interpretations between (30a) and (30b) reduces to whether or not overt (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, and Kastner & Zu (2015) for Latin. What this kind of framework does is to separate the syntactic and semantic behavior of Voice heads, leading to new analyses of argument structure alternations Schäfer 2017).
I take Voice to be instantiated in three different ways, depending on whether it has a [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, either requiring a DP in the specifier of its head, prohibiting a DP in its specifier or remaining agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the structure without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all languages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016;. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two additional templates which we have already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ and hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I explain how it brings about an external argument without requiring one in the syntax (which is what the feature [+D] does, but Voice [+D] will not be discussed further.).
3 Notational matters: the template heXYiZ usually appears in the literature as hiXYiZ, with the first vowel an /i/. Speakers of my generation and at least one generation older use the /e/ form, and so I transcribe /e/ throughout. Conversely, the initial /h/ is usually dropped in speech. I retain it for two reasons. First, /h/ is still pronounced by some older speakers and certain sociolinguistic groups, especially marginalized ones (Schwarzwald 1981;Gafter 2014b is there to force an agentive reading. Doron (2003) proposed that this modifier carries the semantics of Action, which is slightly weaker than that of Agent. I believe that (Kratzer 1996). The typology of Voice heads follows recent proposals by Schäfer (2008) for German, Bruening (2014) for English, Alexiadou & Doron (2012) and Spathas et al. (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, and Kastner & Zu (2015) for Latin. What this kind of framework does is to separate the syntactic and semantic behavior of Voice heads, leading to new analyses of argument structure alternations Schäfer 2017).
I take Voice to be instantiated in three different ways, depending on whether it has a [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, either requiring a DP in the specifier of its head, prohibiting a DP in its specifier or remaining agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the structure without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all languages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016;. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two additional templates which we have already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ and hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I explain how it brings about an external argument without requiring one in the syntax (which is what the feature [+D] does, but Voice [+D] will not be discussed further.).
enforces a reading that has appeared in a number of recent works on argument structure. In their study of animacy in English, Italian, Greek and Russian, Folli & Harley (2008) considered a range of data in which the acceptability of an external argument depends on whether it is teleologically capable of causing the event (as opposed to an agency or animacy restriction). In a study of manner and causation in English, Beavers & Koontz-Garboden (2012) used the notion of actor and non-actor to discuss events in which an animate causer is or is not responsible for the consequences of its act, distinguishing causation from actorhood. In two studies of external arguments in nominalizations, Sichel (2010) and  similarly differentiated agentivity from direct causation. In a study of reflexives in Greek (which we return to in Section 5.2), Spathas et al. (2015) identified the prefix afto-as an anti-assistive modifier, again performing a similar semantic function. And in Tamil, the suffix -koɭ adds affective semantics in a way that is otherwise difficult to pin down immediately (Sundaresan & McFadden 2017).
To be clear, the crosslinguistic claim is not that Reflexive verbs in Hebrew 9 ner 2016), originally proposed as the functional head introducing external arguments (Kratzer 1996). The typology of Voice heads follows recent proposals by Schäfer (2008) for German, Bruening (2014) for English, Alexiadou & Doron (2012) and Spathas et al. (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, and Kastner & Zu (2015) for Latin. What this kind of framework does is to separate the syntactic and semantic behavior of Voice heads, leading to new analyses of argument structure alternations Schäfer 2017). I take Voice to be instantiated in three different ways, depending on whether it has a [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, either requiring a DP in the specifier of its head, prohibiting a DP in its specifier or remaining agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the structure without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all languages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016; is the sole element responsible for all of these cases. Instead, the pretheoretical picture which emerges from these works is that natural language has a way of making this fine-grained distinction, a distinction we are not yet fully able to explain. Since this phenomenon appears to be semantic in nature (rather than demonstrably syntactic), it is formalized in Hebrew using Reflexive verbs in Hebrew ner 2016), originally proposed as the functional head introducing extern (Kratzer 1996). The typology of Voice heads follows recent proposals by S for German, Bruening (2014) for English, Alexiadou & Doron (2012) and (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) and Kastner & Zu (2015) for Latin. What this kind of framework does is t syntactic and semantic behavior of Voice heads, leading to new analyse structure alternations Schäfer 2017).
I take Voice to be instantiated in three different ways, depending on w [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP requiring a DP in the specifier of its head, prohibiting a DP in its specif ing agnostic about the specifier. In the discussion of hitXaY̯ eZ we will m Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leav ture without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with from merging in its specifier. As typically assumed for unaccusative little v or unaccusative V does not assign accusative case itself (Chomsky 1995) or th culation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantia guages. In Hebrew, it has been argued that the three variants have distin (Kastner 2016;. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two additional tem we have already seen in (4)-(6) and which lie at the heart of the current stu hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where it brings about an external argument without requiring one in the syntax ( the feature [+D] does, but Voice [+D] will not be discussed further.).
3 Notational matters: the template heXYiZ usually appears in the literature as hiXYiZ, wit an /i/. Speakers of my generation and at least one generation older use the /e/ form, and s throughout. Conversely, the initial /h/ is usually dropped in speech. I retain it for two rea still pronounced by some older speakers and certain sociolinguistic groups, especially m (Schwarzwald 1981;Gafter 2014b . As a root, this element has phonological and semantic content but no syntactic requirements. Not much hinges on whether this element is a root or a functional head in this language; I take the simple view that it has no syntactic influence, and so is root-like. The question of what other such "underspecified" roots might exist in natural language remains an open one for further crosslinguistic research.
Formalizing this characterization of

Reflexive verbs in Hebrew 9
ner 2016), originally proposed as the functional head introducing external arguments (Kratzer 1996). The typology of Voice heads follows recent proposals by Schäfer (2008) for German, Bruening (2014) for English, Alexiadou & Doron (2012) and Spathas et al. (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, and Kastner & Zu (2015) for Latin. What this kind of framework does is to separate the syntactic and semantic behavior of Voice heads, leading to new analyses of argument structure alternations Schäfer 2017). I take Voice to be instantiated in three different ways, depending on whether it has a [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, either requiring a DP in the specifier of its head, prohibiting a DP in its specifier or remaining agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the structure without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all languages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016; , I assume that it triggers an agentive alloseme of Voice, following Doron (2003Doron ( , 2014, opting not to tread in the murky waters of distinguishing agentivity from "actorhood" and "direct causation". The relevant morphemes have the denotations in (31):

Reflexive verbs in Hebrew 9
ner 2016), originally proposed as the functional head introducing external arguments (Kratzer 1996). The typology of Voice heads follows recent proposals by Schäfer (2008) for German, Bruening (2014) for English, Alexiadou & Doron (2012) and Spathas et al. (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, and Kastner & Zu (2015) for Latin. What this kind of framework does is to separate the syntactic and semantic behavior of Voice heads, leading to new analyses of argument structure alternations Schäfer 2017). I take Voice to be instantiated in three different ways, depending on whether it has a [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, either requiring a DP in the specifier of its head, prohibiting a DP in its specifier or remaining agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the structure without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all languages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016; [+D] will not be discussed further.).
requires, in the semantics, that Voice introduce an Agent rather than a Cause.
3 Notational matters: the template heXYiZ usually appears in the literature as hiXYiZ, with the first vowel an /i/. Speakers of my generation and at least one generation older use the /e/ form, and so I transcribe /e/ throughout. Conversely, the initial /h/ is usually dropped in speech. I retain it for two reasons. First, /h/ is still pronounced by some older speakers and certain sociolinguistic groups, especially marginalized ones (Schwarzwald 1981;Gafter 2014b The tripartite classification of Voice heads is not necessarily instantiated in all languages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016; we have already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ and hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I explain how it brings about an external argument without requiring one in the syntax (which is what the feature [+D] does, but Voice [+D] will not be discussed further.).
3 Notational matters: the template heXYiZ usually appears in the literature as hiXYiZ, with the first vowel an /i/. Speakers of my generation and at least one generation older use the /e/ form, and so I transcribe /e/ throughout. Conversely, the initial /h/ is usually dropped in speech. I retain it for two reasons. First, /h/ is still pronounced by some older speakers and certain sociolinguistic groups, especially marginalized ones (Schwarzwald 1981;Gafter 2014b d. ⟦Voice [-D] ⟧ ↔ λP <s,t> .P The intuition for reflexives, then, is that a construction in which there is only an internal argument, but in which there is also agentive semantics, leads to an interpretation in which the internal argument is also the agent. The next section describes the derivational mechanism which lets one argument receive two thematic roles.

Delayed saturation
The derivational framework adopted in this paper allows for a separation of syntactic and semantic requirements of functional heads. The former hold in the syntax proper, while the latter are computed at LF. This being the case, it is possible for a given syntactic head to impose a semantic requirement that is not fulfilled immediately but only later in the derivation. This possibility, which I call delayed saturation, has been investigated in recent work on argument structure alternations and possession in a number of languages by Wood (2014Wood ( , 2015, Wood & Marantz (2017) and Myler (2016). An example from Icelandic illustrates the basic idea; a formalization for reflexives follows in Section 3.2.3. Icelandic exhibits a specific kind of reflexive-like construction, the "figure reflexive", in which an argument is interpreted both as an agent and as a figure (theme) of a motion event. These constructions appear with the clitic -st, as in (32). On the analysis of Wood (2014) the subject of PPs, the figure, is introduced by a functional head p merging above the PP, following independently made suggestions along similar lines (van Riemsdijk 1990;Rooryck 1996;Koopman 1997;Gehrke 2008;Den Dikken 2003Svenonius 2003Svenonius , 2007Svenonius , 2010. In this system, p assigns the thematic role of figure and Voice assigns agent. These labels indicate semantic interpretation at LF, rather than traditional theta-roles. The structure for (32) is given in (33). Wood's insight is that -st serves as an expletive, filling Spec,pP in the syntax without contributing any semantics. The next DP merged in the structure, Bjartur, will then saturate both Voice's semantic role (agent) and the role of figure introduced by p. A variety of diagnostics show that the verb is agentive, with the DP Bjartur merged in Spec,VoiceP. (33)

Reflexive verbs in Hebrew 17
matic role of figure and Voice assigns agent. These labels indicate semantic interpretation at LF, rather than traditional theta-roles. The structure for (32) is given in (33). Wood's insight is that -st serves as an expletive, filling Spec,pP in the syntax without contributing any semantics. The next DP merged in the structure, Bjartur, will then saturate both Voice's semantic role (agent) and the role of figure introduced by p. A variety of diagnostics show that the verb is agentive, with the DP Bjartur merged in Spec,VoiceP. PP gegnum … The full semantic details can be found in Wood (2014;. The intuition is that a function can remain unsaturated by the syntactic arguments of its head; in this case the semantic role of figure is not saturated by -st, which is the element introduced by p in the syntax. Instead, an argument introduced later on (Bjartur) saturates the predicate.
The full semantic details can be found in Wood (2014Wood ( , 2015. The intuition is that a function can remain unsaturated by the syntactic arguments of its head; in this case the semantic role of figure is not saturated by -st, which is the element introduced by p in the syntax. Instead, an argument introduced later on (Bjartur) saturates the predicate. What I call delayed saturation is more of a side effect of the nature of the derivation than a novel mechanism. It is of course not the norm for saturating functions, since otherwise John kicked could mean 'John i kicked John i ' with delayed saturation of the Agent role.
In what follows I show how delayed saturation is instantiated in the specific environment of Hebrew reflexives.

Putting the pieces together
We are now armed with a non-active Voice head, an agentive modifier and a formalism allowing for an argument to saturate a function lower in the tree. Combining the three should give us an internal argument, which is nevertheless interpreted as an agent. The structure in (35) and the semantic derivation in (36)  The argument DP starts off as the internal argument. No external argument is merged in the specifier of Voice [-D] , but the specifier of T still needs to be filled. The internal argument raises directly to Spec,TP, say to satisfy the EPP, saturating the agent role of Voice [-D] in delayed fashion. Whereas in the Icelandic example it was the figure role whose saturation was delayed until the merger of Spec,TP, here it is Agent(x, e) which cannot be satisfied immediately. The crucial points in this derivation are (36e) and (36g): once the internal argument raises to Spec,TP, the derivation converges.

washed.intns.mid 'Danny washed (himself).'
The argument DP starts off as the internal argument. No external argument is merged in the specifier of Voice [-D] , but the specifier of T still needs to be filled. The internal argument raises directly to Spec,TP, say to satisfy the EPP, saturating the agent role of Voice [-D] in delayed fashion. Whereas in the Icelandic example it was the figure role whose saturation was delayed until the merger of Spec,TP, here it is Agent(x,e) which cannot be satisfied immediately. The crucial points in this derivation are (36e) and (36g) The argument DP starts off as the internal argument. No external argument is merged in the specifier of Voice [-D] , but the specifier of T still needs to be filled. The internal argument raises directly to Spec,TP, say to satisfy the EPP, saturating the agent role of Voice [-D] in delayed fashion. Whereas in the Icelandic example it was the figure role whose saturation was delayed until the merger of Spec,TP, here it is Agent(x,e) which cannot be satisfied immediately. The crucial points in this derivation are (36e) and (36g): once the internal argument raises to Spec,TP, the derivation converges. The argument DP starts off as the internal argument. No external argument is merged in the specifier of Voice [-D] , but the specifier of T still needs to be filled. The internal argument raises directly to Spec,TP, say to satisfy the EPP, saturating the agent role of Voice [-D] in delayed fashion. Whereas in the Icelandic example it was the figure role whose saturation was delayed until the merger of Spec,TP, here it is Agent(x,e) which cannot be satisfied immediately. The crucial points in this derivation are (36e) and (36g): once the internal argument raises to Spec,TP, the derivation converges. ner 2016), originally proposed as the functional head introducing external arguments (Kratzer 1996). The typology of Voice heads follows recent proposals by Schäfer (2008) for German, Bruening (2014) for English, Alexiadou & Doron (2012) and Spathas et al. (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, and Kastner & Zu (2015) for Latin. What this kind of framework does is to separate the syntactic and semantic behavior of Voice heads, leading to new analyses of argument structure alternations Schäfer 2017). I take Voice to be instantiated in three different ways, depending on whether it has a [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, either requiring a DP in the specifier of its head, prohibiting a DP in its specifier or remaining agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the structure without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all languages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016; This analysis showcases what I mean when I correlate complex meaning with complex morphology. On the meaning side of things, reflexives in Hebrew do not come from a dedicated functional or lexical item. There must be some conspiracy of factors in order to derive a reflexive reading. In this, reflexives are different than anticausatives, which can be derived simply by using the head Voice [-D] (Section 2.3). The complex structure of reflexives is tracked by complex morphology: verbs in hitXaY̯ eZ have two distinguishing morphophonological properties, namely the prefix and the non-spirantized medial root consonant Y̯ . We now have answers to the questions posed at the beginning of the paper: why hitXaY̯ eZ is the one template instantiating reflexive verbs, and why this template in particular. Different elements are necessary in order for a reflexive reading to arise, and their combination in the morphophonology results in this template. As returned to in Section 6.2, alternative approaches to Hebrew cannot answer these questions, since they treat each template as an independent morpheme. I use distinct functional heads; Table 1 summarizes the syntactic and semantic contributions of the heads utilized thus far. Empty cells are underspecified. 6 The derivation also required movement from the internal argument position. There is more to be said about this kind of movement; that thread will be picked up in Section 4. Before that can be addressed, we will return once more to anticausatives: at this point, the theory lacks a formalization of why anticausatives and reflexives are different even though their constituent elements are similar. If the morphosyntax is identical, why is the morphosemantics different? The general answer I give is that different classes of roots impose different constraints in the semantics. With Hebrew anticausatives, the requirement on the agent role is relaxed in a way that is made explicit next.

The semantics of anticausatives
The analysis presented here requires hitXaY̯ eZ anticausatives to be built using Reflexive verbs in Hebrew ner 2016), originally proposed as the functional head introd (Kratzer 1996). The typology of Voice heads follows recent p for German, Bruening (2014) for English, Alexiadou & Doron (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, W and Kastner & Zu (2015) for Latin. What this kind of framew syntactic and semantic behavior of Voice heads, leading to structure alternations Schäfer 2017).
I take Voice to be instantiated in three different ways, dep [+D] feature, a [-D] feature or is underspecified for [±D]. Th requiring a DP in the specifier of its head, prohibiting a DP ing agnostic about the specifier. In the discussion of hitXaY Voice [-D] , a Voice head which bans the merger of a DP in its ture without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting a from merging in its specifier. As typically assumed for unaccusative little v or u does not assign accusative case itself (Chomsky culation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessa guages. In Hebrew, it has been argued that the three varian (Kastner 2016; we have already seen in (4)-(6) and which lie at the heart of th hitXaY̯ eZ. 4 This element is discussed in more detail in Sectio it brings about an external argument without requiring one in the feature [+D] does, but Voice [+D] will not be discussed furt 3 Notational matters: the template heXYiZ usually appears in the literatur an /i/. Speakers of my generation and at least one generation older use the throughout. Conversely, the initial /h/ is usually dropped in speech. I reta still pronounced by some older speakers and certain sociolinguistic group (Schwarzwald 1981;Gafter 2014b). Second, the initial h-should help the this template from others. 4 It appears that √ action does not combine with Voice [+D] ; see Kastner ( sion. , accounting for their morphophonological form. But this modifier cannot do its regular semantic work, otherwise we would expect an agent for anticausatives, contrary to fact. 6 The combination of Voice and

Reflexive verbs in Hebrew 9
ner 2016), originally proposed as the functional head introducing external arguments (Kratzer 1996). The typology of Voice heads follows recent proposals by Schäfer (2008) for German, Bruening (2014) for English, Alexiadou & Doron (2012) and Spathas et al. (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, and Kastner & Zu (2015) for Latin. What this kind of framework does is to separate the syntactic and semantic behavior of Voice heads, leading to new analyses of argument structure alternations Schäfer 2017). I take Voice to be instantiated in three different ways, depending on whether it has a [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, either requiring a DP in the specifier of its head, prohibiting a DP in its specifier or remaining agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the structure without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all languages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016; [+D] will not be discussed further.).
3 Notational matters: the template heXYiZ usually appears in the literature as hiXYiZ, with the first vowel an /i/. Speakers of my generation and at least one generation older use the /e/ form, and so I transcribe /e/ throughout. Conversely, the initial /h/ is usually dropped in speech. I retain it for two reasons. First, /h/ is still pronounced by some older speakers and certain sociolinguistic groups, especially marginalized ones (Schwarzwald 1981;Gafter 2014b). Second, the initial h-should help the non-Semitist reader distinguish this template from others. 4 It appears that √ action does not combine with Voice [+D] ; see Kastner (2016: Chapter 2.4.1) for discussion.
cannot lead to a reflexive verb since external Merge would generate a subject from the numeration in Spec,Voice ("Merge over Move"). Alternatively, Oseki (2017)   ner 2016), originally proposed as the functional head introducing external arguments (Kratzer 1996). The typology of Voice heads follows recent proposals by Schäfer (2008) for German, Bruening (2014) for English, Alexiadou & Doron (2012) and Spathas et al. (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, and Kastner & Zu (2015) for Latin. What this kind of framework does is to separate the syntactic and semantic behavior of Voice heads, leading to new analyses of argument structure alternations Schäfer 2017). I take Voice to be instantiated in three different ways, depending on whether it has a [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, either requiring a DP in the specifier of its head, prohibiting a DP in its specifier or remaining agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the structure without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all languages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016;. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two additional templates which we have already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ and hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I explain how it brings about an external argument without requiring one in the syntax (which is what the feature [+D] does, but Voice [+D] will not be discussed further.).
3 Notational matters: the template heXYiZ usually appears in the literature as hiXYiZ, with the first vowel an /i/. Speakers of my generation and at least one generation older use the /e/ form, and so I transcribe /e/ throughout. Conversely, the initial /h/ is usually dropped in speech. I retain it for two reasons. First, /h/ is still pronounced by some older speakers and certain sociolinguistic groups, especially marginalized ones (Schwarzwald 1981;Gafter 2014b

XiY̯ eZ
Voice [+D] Obligatory external argument he-XYiZ Voice [-D] No external argument ni-XYaZ Voice [-D] Reflexive verbs in Hebrew 9 ner 2016), originally proposed as the functional head introducing external arguments (Kratzer 1996). The typology of Voice heads follows recent proposals by Schäfer (2008) for German, Bruening (2014) for English, Alexiadou & Doron (2012) and Spathas et al. (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, and Kastner & Zu (2015) for Latin. What this kind of framework does is to separate the syntactic and semantic behavior of Voice heads, leading to new analyses of argument structure alternations Schäfer 2017). I take Voice to be instantiated in three different ways, depending on whether it has a [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, either requiring a DP in the specifier of its head, prohibiting a DP in its specifier or remaining agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the structure without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all languages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016;. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two additional templates which we have already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ and hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I explain how it brings about an external argument without requiring one in the syntax (which is what the feature [+D] does, but Voice [+D] will not be discussed further.).

I propose that the rule of allosemy (semantic interpretation) in (37) removes the agentivity requirement of
The tripartite classification of Voice heads is not necessarily instantiated in all lan-guages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016; [+D] will not be discussed further.).
3 Notational matters: the template heXYiZ usually appears in the literature as hiXYiZ, with the first vowel an /i/. Speakers of my generation and at least one generation older use the /e/ form, and so I transcribe /e/ throughout. Conversely, the initial /h/ is usually dropped in speech. I retain it for two reasons. First, /h/ is still pronounced by some older speakers and certain sociolinguistic groups, especially marginalized ones (Schwarzwald 1981;Gafter 2014b f. √ bʃl biʃel 'cooked' hitbaʃel 'got cooked' In all these cases, the non-active version is a detransitivized form of the active version and shares the same root as the active verb. The derived verbs in (10) are all intransitive and their bases transitive.

Agent-oriented adverbs
A number of agent-oriented tests show that no agents are possible with anticausatives in Hebrew. The phrase me-aʦmo 'by itself' diagnoses the lack of an external argument (Levin & Rappaport Hovav 1995;Alexiadou & Anagnostopoulou 2004;Alexiadou & Doron 2012), be it explicit (as with transitive verbs) or implicit (as with passives). The test is appropriate with anticausatives in the niXYaZ template shown above, (11a), but not with direct objects of transitive verbs, (11b), or with passive verbs, (11c-d). , seen for example in (24). This change renders the resulting verb hitparek 'fell apart' anticausative, rather than a potential reflexive 'tore himself to pieces'. The process can be thought of as similar to impoverishment (Bonet 1991;Noyer 1998) but in the semantics (Nevins 2015). 7 (37) ⟦ As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991).
. Voice [-D] = λP <s,t> .P tripartite classification of Voice heads is not necessarily instantiated in all lan-In Hebrew, it has been argued that the three variants have distinct exponents r 2016; 2017). .
Voice [+D] ↔ heXYiZ 3 ing an agentive modifier, √ action, derives the two additional templates which already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ and Z. 4 This element is discussed in more detail in Section 3.2, where I explain how s about an external argument without requiring one in the syntax (which is what ure [+D] does, but Voice [+D] will not be discussed further. hitXaY̯ eZ, with other templates brought up as necessary. Each template is traditionally associated with a certain kind of argument structure alternation: causative, passive, and so on (Berman 1978;Doron 2003;Arad 2005;Borer 2013;Kastner 2016 'The gate opened.' Simple as this alternation may be, it does not generalize to all roots. First, there exist anticausative verbs in niXYaZ, (2), with no causative alternation in XaYaZ from which they could have been derived. While causative variants exist in other templates, it is debatable whether a strict derivational relationship should be postulated as in e.g. Laks (2013). Second, there exist verbs in niXYaZ which are not anticausative, (3): they can be shown to pattern with unergative verbs, rather than unaccusative ones (Kastner 2016). In addition, not all roots instantiate verbs in all seven templates, indicating a high degree of lexical idiosyncrasy: for each root, information must be listed indicating which template it can appear in.
(3) gilad Gilad nixnas entered.mid be-gaava in-pride la-bajt to.the-house ha-xadaʃ. the-new 'Gilad entered his new house proudly.' Of interest in the current paper is the "complex" template hitXaY̯ eZ, exemplified in (4)-(6). The typical alternation for this template is an anticausative one, between a transitive verb in XiY̯ eZ and an anticausative in hitXaY̯ eZ (the notation Y̯ indicates lack of spirantization, a phonological process I return to in Section 3). We will focus on the fact that verbs in hitXaY̯ eZ can have other readings associated with them, (5)-(6), besides an anticausative correspondent of XiY̯ eZ.

Reflexive verbs in Hebrew 3
hitXaY̯ eZ, with other templates brought up as necessary. Each template is traditionally associated with a certain kind of argument structure alternation: causative, passive, and so on (Berman 1978;Doron 2003;Arad 2005;Borer 2013;Kastner 2016 'The gate opened.' Simple as this alternation may be, it does not generalize to all roots. First, there exist anticausative verbs in niXYaZ, (2), with no causative alternation in XaYaZ from which they could have been derived. While causative variants exist in other templates, it is debatable whether a strict derivational relationship should be postulated as in e.g. Laks (2013). Second, there exist verbs in niXYaZ which are not anticausative, (3): they can be shown to pattern with unergative verbs, rather than unaccusative ones (Kastner 2016). In addition, not all roots instantiate verbs in all seven templates, indicating a high degree of lexical idiosyncrasy: for each root, information must be listed indicating which template it can appear in.
(3) gilad Gilad nixnas entered.mid be-gaava in-pride la-bajt to.the-house ha-xadaʃ. the-new 'Gilad entered his new house proudly.' Of interest in the current paper is the "complex" template hitXaY̯ eZ, exemplified in (4)-(6). The typical alternation for this template is an anticausative one, between a transitive verb in XiY̯ eZ and an anticausative in hitXaY̯ eZ (the notation Y̯ indicates lack of spirantization, a phonological process I return to in Section 3). We will focus on the fact that verbs in hitXaY̯ eZ can have other readings associated with them, (5)-(6), besides an anticausative correspondent of XiY̯ eZ.

Anticausatives in Hebrew
The traditional view of Semitic templates is that they encode argument structure a nations, as in (1) and (4)  got In all these cases, the non-active version is a detransitivized form of the active sion and shares the same root as the active verb. The derived verbs in (10) are all int sitive and their bases transitive.

Agent-oriented adverbs
ner 2016), originally proposed as the functional head introducing external arguments (Kratzer 1996). The typology of Voice heads follows recent proposals by Schäfer (2008) for German, Bruening (2014) for English, Alexiadou & Doron (2012) and Spathas et al. (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, and Kastner & Zu (2015) for Latin. What this kind of framework does is to separate the syntactic and semantic behavior of Voice heads, leading to new analyses of argument structure alternations Schäfer 2017).
I take Voice to be instantiated in three different ways, depending on whether it has a [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, either requiring a DP in the specifier of its head, prohibiting a DP in its specifier or remaining agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the structure without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all languages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016;. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two additional templates which we have already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ and hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I explain how it brings about an external argument without requiring one in the syntax (which is what the feature [+D] does, but Voice [+D] will not be discussed further.).
3 Notational matters: the template heXYiZ usually appears in the literature as hiXYiZ, with the first vowel an /i/. Speakers of my generation and at least one generation older use the /e/ form, and so I transcribe /e/ throughout. Conversely, the initial /h/ is usually dropped in speech. I retain it for two reasons. First, /h/ is still pronounced by some older speakers and certain sociolinguistic groups, especially marginalized ones (Schwarzwald 1981;Gafter 2014b ), used elsewhere in the language; and a general compositional mechanism of delayed saturation.
It was also shown that this approach explains why the same morphology might signal different syntactic derivations. A correlation was identified between complex syntax/semantics and complex morphophonology: on the present theory, reflexive readings do not come as primitives but arise as the result of specific heads combining. Since each of these heads also has its own exponent, the marked syntax/semantics is reflected in marked morphology, thereby explaining why it is this specific morphology (the template hitXaY̯ eZ) which is used for this specific kind of verb (reflexive).
In addition, the analysis supported a division of roots into two kinds which can be distinguished on lexical semantic grounds. We have also scratched the surface of restrictions and triggers of A-movement. These two issues are expounded on next: how syntactic structure, namely unaccusativity, interacts with the root's own characteristics.

Unaccusativity and lexical semantics
With the anaylsis of reflexives and anticausatives under our belt, we explore next the broader implications for the theoretical architecture defended here: deep and surface unaccusativity (in Section 4.1) and the role of roots in the derivation (in Section 4.2).

Deep and surface unaccusativity
My analysis of reflexive verbs in Hebrew treats them as unaccusative, although I have not shown whether they pass unaccusativity diagnostics. They do not: VS order #hitkalx-u ʃloʃa xatulim be-arba ba-boker. showered.intns.mid-3pl three cats in-four in.the-morning (int. 'Three cats washed themselves at 4am.') Possessive dative #ʃloʃa xatulim hitkalx-u l-i be-arba ba-boker three cats showered.intns.mid-3pl to-me in-four in.the-morning 'Three cats washed themselves at 4am and I was adversely affected.' (# int. 'My three cats washed themselves at 4am.') In this section I revisit these diagnostics, asking what it is exactly that they diagnose. Examination of VS order, in particular, reveals that it is not always useful to speak of "unaccusativity" as a holistic concept. Instead, what matters is where arguments are generated and where they end up in the course of the derivation.

Verb-Subject order
VS order is not possible with reflexives, (38). However, we should ask what the diagnostic is actually diagnosing. In the analysis of reflexives proposed here the internal argument undergoes A-movement to Spec,TP and ends up higher than its base-generated position, as in (35) above.
It is likely that VS order only diagnoses surface unaccusativity, that is, a structure in which the internal argument remains in its base-generated position, rather than deep unaccusativity. The difference between the two was most clearly noted by Levin & Rappaport Hovav (1995). It has been proposed that the subjects of "deep" unaccusatives originated as internal arguments but have moved to subject position, while "surface" unaccusatives remain in their low, base-generated position, (40). Viewed in these terms, Italian ne-cliticization (Burzio 1986) is a surface diagnostic. The internal argument can either stay in its base-generated position, (41a), or raise, (41b). But the object out of which the clitic ne 'of them' is extracted must remain in its base-generated position, (42). See Burzio (1986: 23) and Irwin (2012: 32) for additional discussion.
(41) Italian a. Baseline example, internal argument remains low Saranno invitati [molti esperti]. will.be invited many experts 'Many experts will be invited.' b. Baseline example, internal argument raises [Molti esperti] saranno invitati _____ . many experts will.be invited 'Many experts will be invited.' (=a) (42) Italian a. Ne-cliticization allowed out of a surface object Ne saranno invitati [molti _____ ]. of.them will.be invited many 'Many of them will be invited.' b. *Ne-cliticization disallowed out of a moved, "deep" object [Molti ____ ] ne saranno invitati. many of.them will.be invited (int. 'Many of them will be invited.') Here is what is at stake: if VS order in Hebrew is a "surface" unaccusativity diagnostic, then this would explain why reflexives do not pass it -the internal argument has moved out of the VP and into subject position. Unfortunately, there is little additional evidence for or against the claim that VS order in Hebrew is a "surface" unaccusativity diagnostic. Instead, we must leave this as a conjecture to be explored in a related line of inquiry: why can Hebrew anticausative arguments remain low and ignore the EPP?
The word order facts introduced in Section 2.2 indicate that an anticausative object may either stay low or raise to Spec,TP. But the reflexive internal argument must raise if the derivation is to converge; if it does not, no argument satisfies the Agent role and the derivation crashes at the interface with LF.
I have not given an explicit account of the optionality of movement for anticausative arguments, which unlike reflexive arguments are allowed to stay low. This, I believe, is a challenge for all research on unaccusativity. As seen in (41a-b), the internal argument in Italian may either stay low or raise, with no apparent difference in interpretation.
A number of open questions remain: why do Italian and Hebrew allow for this "optional" movement, allowing unaccusatives to remain low? If the EPP forces movement to Spec,TP, can it be "turned off" or satisfied in another way (Alexiadou & Anagnostopoulou 1998)? The answers to these questions lie beyond the scope of the current account. But when similar questions have been tackled, the resulting accounts suggest that VS order is not necessarily about unaccusativity per se, but about a certain syntactic configuration that has particular semantic and information-structural consequences as well, in line with the analysis advanced here (Borer 2005;Alexiadou 2011). It is my hope that the phenomena investigated in the current paper can serve as a stepping stone for further work on this topic.

Possessive dative
The other diagnostic proposed in the literature on Hebrew is the possessive dative, which has recently been re-characterized by Gafter (2014a) and Linzen (2014Linzen ( , 2016 as a diagnostic of saliency or animacy rather than unaccusativity. Gafter (2014a) gives the following contrast by way of example: (43) a. ha-karborator neheras le-dan. the-carburetor ruined.mid to-Dan 'Dan's carburetor got ruined.' b. *ha-karborator neheras la-mexonit. the-carburetor ruined.mid to.the-car (int. 'The car's carburetor got ruined.') The animate possessor in (43a) is acceptable, but the inanimate possessor in (43b) is not. Taking these kinds of data as his point of departure, Gafter (2014a) conducted a rating study to test whether the prominence of the possessor was the crucial factor driving grammaticality in the possessive dative, where prominence is defined both in terms of animacy and definiteness. The experiment bore out this prediction.
In a reflexive construction such as that in (39), the to-be-possessed argument ('cats') is animate since it is the agent of a reflexive predicate. As Gafter shows, this is a case where acceptability of possessive datives suffers when both possessor and possessee are animate and salient in the discourse.
A prediction made by this account is that a 3rd person possessive dative should not be possible with a 1st person possessee. 9 This seems to be correct: (44) *niftsa-ti la-kvutsa. injured.mid-1sg to.the-team (int. 'I got injured, and I was part of the team.') On the one hand, these findings provide us with an out by denying the applicability of the diagnostic. If the possessive dative is not really an unaccusativity diagnostic, then the fact that reflexives do not pass it does not argue against an unaccusative analysis. On the other hand, this failure to pass the diagnostic may be interesting in its own right. As a first pass, it shows that affectedness has a number of syntactic as well as semantic causes.

Unaccusative and unergative reflexives
To summarize the discussion of these two unaccusativity diagnostics, I have argued that the broad notion of "unaccusativity" is not enough to describe reflexives in Hebrew (and is too broad in general for other phenomena; Irwin 2012; Alexiadou 2011Alexiadou , 2014. A similar idea will be necessary for the discussion of Greek in Section 5.2. If unaccusativity means that the surface subject started off as the internal argument, then surface unaccusativity diagnostics might not identify reflexive structures in which the internal argument raised to subject. An anonymous reviewer asks whether there are verbal constructions that contain only VoiceP, in which case the internal argument of reflexives cannot raise to Spec,TP. Unfortunately, the relevant constructions do not deliver clear results in Hebrew. Infinitives have a marked morphological form, presumably the spell-out of non-finite T, e.g. le-hitlabeʃ 'to-get.dressed'. The next candidate is nominalizations, but it is well-known that these can trigger existential closure over the external argument (Grimshaw 1990;Bruening 2013: 31): the Agent is not overtly named in The destruction of the city.
Granted, with no appropriate tests for deep unaccusativity in Hebrew, the idea that reflexives are unaccusative remains a working hypothesis to be explored rather than a conclusion based on established diagnostics. Nevertheless, semantically the argument of reflexive verbs does behave like an internal argument in that it undergoes change of state: if Dina shaves herself, she is now in a shaven state. If John applies make-up to himself, he is now made-up. This behavior is typical of internal arguments (Dowty 1991;Alexiadou & Schäfer 2013).
The debate on whether reflexives are unaccusative or unergative goes back at least to Kayne (1975) and Marantz (1984); see Chierchia (2004), Rappaport Hovav (2009) andSportiche (2014) for recent contrasting views. The answer may vary by language, depending on how a given language promotes its internal arguments. What I have suggested here is that minimal differences between deep and surface unaccusatives might be findable in other languages, even if they are not obvious in Hebrew.

The right root in the right place
The final issue to be raised before evaluating alternative theories is the one relating to the difference between reflexives and anticausatives. In this section I address the question of which roots can be embedded in different contexts: if root A derives a reflexive verb and root B an anticausative one, is it necessary to postulate different derivations or would it be simpler to adopt a lexicalist notion in which each verb projects its own argument structure?
Recent work on argument structure has seen a spate of analyses proposing distinctions between different kinds of roots; see the ontologies proposed by Anagnostopoulou & Samioti (2014) and Levinson (2014), for example. Following Alexiadou (2015), I have made a distinction between Self-Oriented roots and Other-Oriented roots (Section 3.3). 10 These are not syntactic notions but semantic ones, and their purpose is to give us tools with which to discuss different interpretations of verbal structures. The emerging picture for Hebrew is presented in Table 2, which summarizes the different readings that emerge in hitXaY̯ eZ. Reflexives and anticausatives were the subject of the current paper. The framework allows for similar analyses of other verbs in the same template, such as the reciprocals noted earlier on in (4)-(6), but reciprocals themselves will not be dealt with here; it has been argued by Bar-Asher Siegal (2016) that reciprocalization in Hebrew is tangential to the choice of template, since the same reciprocalization strategy (e.g. a plural subject) is possible in a number of templates. I will tentatively assume that a unified analysis of reciprocals in Hebrew would pick out a subset of templates, and not a unique one like with reflexives and hitXaY̯ eZ.
In anticipation of future work, I would like to ask how deterministic these readings are. Compare

Anticausatives in Hebrew
ditional view of Semitic templates is that they encode argument structure alters, as in (1) and (4)  XaYaZ ∼ niXYaZ hitbaʃel 'got cooked' all these cases, the non-active version is a detransitivized form of the active verd shares the same root as the active verb. The derived verbs in (10) are all intrannd their bases transitive.

Agent-oriented adverbs
ber of agent-oriented tests show that no agents are possible with anticausatives rew. The phrase me-aʦmo 'by itself' diagnoses the lack of an external argument & Rappaport Hovav 1995; Alexiadou & Anagnostopoulou 2004;Alexiadou & 2012), be it explicit (as with transitive verbs) or implicit (as with passives). The appropriate with anticausatives in the niXYaZ template shown above, (11a), but th direct objects of transitive verbs, (11b), or with passive verbs, (11c-d). a.
ha-kise the-chair But Hebrew hitXaY̯ eZ is unambiguous. The verb hitlabeʃ 'got dressed' is deterministically reflexive and cannot be used in an anticausative (or reciprocal) context, as shown by its incompatibility with 'by itself' in (28a). In contrast, the verb hitaʦben 'got annoyed' is uniformly anticausative and cannot be used with an agent-oriented adverb such as 'on purpose' (Alexiadou & Anagnostopoulou 2004)  to.him 'The player got annoyed when he wasn't passed the ball.' I argue below that this contrast ultimately derives from the lexical semantics of the root. Dressing up is usually something one does on oneself, while annoying is usually something that one does to someone else. This notion will be made precise in Section 3.3. For now, note that the root itself is not enough to force a reflexive reading. The root √ lbʃ from (28a) can appear in other templates with non-reflexive (and nonanticausative) meanings. Both examples in (29)  the-king 'The tailors dressed up the king.' The point is once again that it is not enough for the root to be compatible with a reflexive reading in order for the verb to be reflexive. In English, for instance, wash and shave do not require any special morphological marking. But in Hebrew, both the root and the template combine to decide the meaning and argument structure of the verb, as I explain next.

Analysis
The intuition behind the analysis is as follows: reflexive verbs in hitXaY̯ eZ consist of an unaccusative structure with extra agentive semantics. This combination is only possible if the internal argument is allowed to saturate the semantic function of an external argument, in a way I formalize below.
wear: the former gives rise to anticausative hitpoʦeʦ and the latter to reflexive hitlabeʃ.

Anticausatives in Hebrew
traditional view of Semitic templates is that they encode argument structure alterions, as in (1) and (4)  XaYaZ ∼ niXYaZ hitbaʃel 'got cooked' In all these cases, the non-active version is a detransitivized form of the active verand shares the same root as the active verb. The derived verbs in (10)  Other-Oriented hitpoʦeʦ 'exploded' (anticausative) b.

Reflexive verbs in Hebrew 13
But Hebrew hitXaY̯ eZ is unambiguous. The verb hitlabeʃ 'got dressed' is deterministically reflexive and cannot be used in an anticausative (or reciprocal) context, as shown by its incompatibility with 'by itself' in (28a). In contrast, the verb hitaʦben 'got annoyed' is uniformly anticausative and cannot be used with an agent-oriented adverb such as 'on purpose' (Alexiadou & Anagnostopoulou 2004)  to.him 'The player got annoyed when he wasn't passed the ball.' I argue below that this contrast ultimately derives from the lexical semantics of the root. Dressing up is usually something one does on oneself, while annoying is usually something that one does to someone else. This notion will be made precise in Section 3.3. For now, note that the root itself is not enough to force a reflexive reading. The point is once again that it is not enough for the root to be compatible with a reflexive reading in order for the verb to be reflexive. In English, for instance, wash and shave do not require any special morphological marking. But in Hebrew, both the root and the template combine to decide the meaning and argument structure of the verb, as I explain next.

Analysis
The intuition behind the analysis is as follows: reflexive verbs in hitXaY̯ eZ consist of an unaccusative structure with extra agentive semantics. This combination is only possible if the internal argument is allowed to saturate the semantic function of an external argument, in a way I formalize below.

Self-Oriented
hitlabeʃ 'dressed up' (reflexive) Interestingly, some Other-Oriented roots can be treated as Self-Oriented in the right context, (46), but Self-Oriented roots cannot be interpreted as Other-Oriented, (47).
(46) Other-Oriented 26 Kastner with here; it has been argued by Bar-Asher Siegal (2016) that reciprocalization in Hebrew is tangential to the choice of template, since the same reciprocalization strategy (e.g. a plural subject) is possible in a number of templates. I will tentatively assume that a unified analysis of reciprocals in Hebrew would pick out a subset of templates, and not a unique one like with reflexives and hitXaY̯ eZ.
An anonymous reviewer similarly claims that the verb hitnaka 'got himself clean' is ambiguous between an anticausative reading, (48a), and a reflexive reading (see Doron 2003: 11 for a similar claim). Perhaps the crucial factor here is the type of event, interacting with the animacy of the subject, i.e. the internal argument: (48b) is only natural with the adverbial and purpose clause.
in a reflexive context, licit le-marbe ha-mazal, ha-mexabel ha-mitabed hitpoʦeʦ be-migraʃ rek. to-much the-luck, the-terrorist the-suiciding exploded.intns.mid in-lot empty 'Luckily, the suicide bomber blew himself up in an empty lot.' (47) Self-Oriented 26 Kastner with here; it has been argued by Bar-Asher Siegal (2016) that reciprocalization in Hebrew is tangential to the choice of template, since the same reciprocalization strategy (e.g. a plural subject) is possible in a number of templates. I will tentatively assume that a unified analysis of reciprocals in Hebrew would pick out a subset of templates, and not a unique one like with reflexives and hitXaY̯ eZ.
An anonymous reviewer similarly claims that the verb hitnaka 'got himself clean' is ambiguous between an anticausative reading, (48a), and a reflexive reading (see Doron 2003: 11 for a similar claim). Perhaps the crucial factor here is the type of event, interacting with the animacy of the subject, i.e. the internal argument: (48b) is only natural with the adverbial and purpose clause.
in a disjoint context, illicit 'The king was still in his underwear minutes before the ceremony. His assistants rushed to dress him up in expensive clothes, a robe and a crown. … *lifnej ʃe-hu hevin ma kara hu kvar hitlabeʃ. before comp-he understood.caus what happened he already dressed.up.intns.mid (… 'before he could understand what had happened, he had already dressed up.)' 10 Alexiadou (2015) actually suggested a tripartite division based mostly on Dutch, in which some roots are inherently reflexive (e.g.

Reflexive verbs in Hebrew 25
Unfortunately, the relevant constructions do not deliver clear results in Hebrew. Infinitives have a marked morphological form, presumably the spell-out of non-finite T, e.g. lehitlabeʃ 'to-get.dressed' . The next candidate is nominalizations, but it is well-known that these can trigger existential closure over the external argument (Grimshaw 1990;Bruening 2013: 31): the Agent is not overtly named in The destruction of the city. Granted, with no appropriate tests for deep unaccusativity in Hebrew, the idea that reflexives are unaccusative remains a working hypothesis to be explored rather than a conclusion based on established diagnostics. Nevertheless, semantically the argument of reflexive verbs does behave like an internal argument in that it undergoes change of state: if Dina shaves herself, she is now in a shaven state. If John applies make-up to himself, he is now made-up. This behavior is typical of internal arguments (Dowty 1991;Alexiadou & Schäfer 2013).
The debate on whether reflexives are unaccusative or unergative goes back at least to Kayne (1975) and Marantz (1984); see Chierchia (2004), Rappaport Hovav (2009) andSportiche (2014) for recent contrasting views. The answer may vary by language, depending on how a given language promotes its internal arguments. What I have suggested here is that minimal differences between deep and surface unaccusatives might be findable in other languages, even if they are not obvious in Hebrew.

The right root in the right place
The final issue to be raised before evaluating alternative theories is the one relating to the difference between reflexives and anticausatives. In this section I address the question of which roots can be embedded in different contexts: if root A derives a reflexive verb and root B an anticausative one, is it necessary to postulate different derivations or would it be simpler to adopt a lexicalist notion in which each verb projects its own argument structure?
Recent work on argument structure has seen a spate of analyses proposing distinctions between different kinds of roots; see the ontologies proposed by Anagnostopoulou & Samioti (2014) and Levinson (2014), for example. Following Alexiadou (2015), I have made a distinction between Self-Oriented roots and Other-Oriented roots (Section 3.3). 10 These are not syntactic notions but semantic ones, and their purpose is to give us tools with which to discuss different interpretations of verbal structures. The emerging picture for Hebrew is presented in Table 2, which summarizes the different readings that emerge in hitXaY̯ eZ. Reflexives and anticausatives were the subject of the current paper. The framework allows for similar analyses of other verbs in the same template, such as the reciprocals noted earlier on in (4)-(6), but reciprocals themselves will not be dealt 10 Alexiadou (2015) actually suggested a tripartite division based mostly on Dutch, in which some roots are inherently reflexive (e.g. √ shame), some naturally reflexive/reciprocal (e.g. √ wash) and some naturally disjoint (e.g. √ hate). I will make do with a binary distinction.

Reflexive verbs in Hebrew 25
Unfortunately, the relevant constructions do not deliver clear results in Hebrew. Infinitives have a marked morphological form, presumably the spell-out of non-finite T, e.g. lehitlabeʃ 'to-get.dressed' . The next candidate is nominalizations, but it is well-known that these can trigger existential closure over the external argument (Grimshaw 1990;Bruening 2013: 31): the Agent is not overtly named in The destruction of the city. Granted, with no appropriate tests for deep unaccusativity in Hebrew, the idea that reflexives are unaccusative remains a working hypothesis to be explored rather than a conclusion based on established diagnostics. Nevertheless, semantically the argument of reflexive verbs does behave like an internal argument in that it undergoes change of state: if Dina shaves herself, she is now in a shaven state. If John applies make-up to himself, he is now made-up. This behavior is typical of internal arguments (Dowty 1991;Alexiadou & Schäfer 2013).
The debate on whether reflexives are unaccusative or unergative goes back at least to Kayne (1975) and Marantz (1984); see Chierchia (2004), Rappaport Hovav (2009) andSportiche (2014) for recent contrasting views. The answer may vary by language, depending on how a given language promotes its internal arguments. What I have suggested here is that minimal differences between deep and surface unaccusatives might be findable in other languages, even if they are not obvious in Hebrew.

The right root in the right place
The final issue to be raised before evaluating alternative theories is the one relating to the difference between reflexives and anticausatives. In this section I address the question of which roots can be embedded in different contexts: if root A derives a reflexive verb and root B an anticausative one, is it necessary to postulate different derivations or would it be simpler to adopt a lexicalist notion in which each verb projects its own argument structure?
Recent work on argument structure has seen a spate of analyses proposing distinctions between different kinds of roots; see the ontologies proposed by Anagnostopoulou & Samioti (2014) and Levinson (2014), for example. Following Alexiadou (2015), I have made a distinction between Self-Oriented roots and Other-Oriented roots (Section 3.3). 10 These are not syntactic notions but semantic ones, and their purpose is to give us tools with which to discuss different interpretations of verbal structures. The emerging picture for Hebrew is presented in Table 2, which summarizes the different readings that emerge in hitXaY̯ eZ. Reflexives and anticausatives were the subject of the current paper. The framework allows for similar analyses of other verbs in the same template, such as the reciprocals noted earlier on in (4)-(6), but reciprocals themselves will not be dealt 10 Alexiadou (2015) actually suggested a tripartite division based mostly on Dutch, in which some roots are inherently reflexive (e.g. √ shame), some naturally reflexive/reciprocal (e.g. √ wash) and some naturally disjoint (e.g. √ hate). I will make do with a binary distinction.

Reflexive verbs in Hebrew 25
Unfortunately, the relevant constructions do not deliver clear results in Hebrew. Infinitives have a marked morphological form, presumably the spell-out of non-finite T, e.g. lehitlabeʃ 'to-get.dressed' . The next candidate is nominalizations, but it is well-known that these can trigger existential closure over the external argument (Grimshaw 1990;Bruening 2013: 31): the Agent is not overtly named in The destruction of the city. Granted, with no appropriate tests for deep unaccusativity in Hebrew, the idea that reflexives are unaccusative remains a working hypothesis to be explored rather than a conclusion based on established diagnostics. Nevertheless, semantically the argument of reflexive verbs does behave like an internal argument in that it undergoes change of state: if Dina shaves herself, she is now in a shaven state. If John applies make-up to himself, he is now made-up. This behavior is typical of internal arguments (Dowty 1991;Alexiadou & Schäfer 2013).
The debate on whether reflexives are unaccusative or unergative goes back at least to Kayne (1975) and Marantz (1984); see Chierchia (2004), Rappaport Hovav (2009) andSportiche (2014) for recent contrasting views. The answer may vary by language, depending on how a given language promotes its internal arguments. What I have suggested here is that minimal differences between deep and surface unaccusatives might be findable in other languages, even if they are not obvious in Hebrew.

The right root in the right place
The final issue to be raised before evaluating alternative theories is the one relating to the difference between reflexives and anticausatives. In this section I address the question of which roots can be embedded in different contexts: if root A derives a reflexive verb and root B an anticausative one, is it necessary to postulate different derivations or would it be simpler to adopt a lexicalist notion in which each verb projects its own argument structure?
Recent work on argument structure has seen a spate of analyses proposing distinctions between different kinds of roots; see the ontologies proposed by Anagnostopoulou & Samioti (2014) and Levinson (2014), for example. Following Alexiadou (2015), I have made a distinction between Self-Oriented roots and Other-Oriented roots (Section 3.3). 10 These are not syntactic notions but semantic ones, and their purpose is to give us tools with which to discuss different interpretations of verbal structures. The emerging picture for Hebrew is presented in Table 2, which summarizes the different readings that emerge in hitXaY̯ eZ. Reflexives and anticausatives were the subject of the current paper. The framework allows for similar analyses of other verbs in the same template, such as the reciprocals noted earlier on in (4)-(6), but reciprocals themselves will not be dealt 10 Alexiadou (2015) actually suggested a tripartite division based mostly on Dutch, in which some roots are inherently reflexive (e.g. √ shame), some naturally reflexive/reciprocal (e.g. √ wash) and some naturally disjoint (e.g. √ hate). I will make do with a binary distinction.
). I will make do with a binary distinction.

Self-Oriented root Other-Oriented root …
Reflexive verbs in Hebrew 9 ner 2016), originally proposed as the functional head introducing external arguments (Kratzer 1996). The typology of Voice heads follows recent proposals by Schäfer (2008) for German, Bruening (2014) for English, Alexiadou & Doron (2012) and Spathas et al. (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, and Kastner & Zu (2015) for Latin. What this kind of framework does is to separate the syntactic and semantic behavior of Voice heads, leading to new analyses of argument structure alternations Schäfer 2017). I take Voice to be instantiated in three different ways, depending on whether it has a [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, either requiring a DP in the specifier of its head, prohibiting a DP in its specifier or remaining agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the structure without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all languages. In Hebrew, it has been argued that the three variants have distinct exponents (Kastner 2016; [+D] will not be discussed further.).
b. ha-jeladim i hexmiʦ-u (*pro i ). the-children missed.out.caus-3pl 'The children missed.' (Not possible: *'The children missed themselves') Something special would have to be said about Voice [-D] and culation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all guages. In Hebrew, it has been argued that the three variants have distinct expon (Kastner 2016; we have already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I explain it brings about an external argument without requiring one in the syntax (which is w the feature [+D] does, but Voice [+D] will not be discussed further.).
3 Notational matters: the template heXYiZ usually appears in the literature as hiXYiZ, with the first v an /i/. Speakers of my generation and at least one generation older use the /e/ form, and so I transcrib throughout. Conversely, the initial /h/ is usually dropped in speech. I retain it for two reasons. First, still pronounced by some older speakers and certain sociolinguistic groups, especially marginalized (Schwarzwald 1981;Gafter 2014b , i.e. about the morphology of hitXaY̯ eZ, whereas my account strives for compositionality across templates. On the account defended in this paper, the syntax, semantics and phonology of both Voice [-D] and from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991 are consistent in different derivations. There is one choice point at which a root-specific interpretation applies in order to distinguish inchoatives from reflexives, but otherwise the system proceeds as expected. Second, the meaning of a reflexive verb in hitXaY̯ eZ is not the same as a transitive verb in XiY̯ eZ with a reflexive anaphor, a fact which casts doubt on the silent anaphor analysis. Let us see why this is so. As shown by Doron & Rappaport Hovav (2009: 93), a syntactic reflexive with an anaphor, (50a), differs in meaning from a morphological reflexive, While this fact does not falsify an anaphor-based analysis, it does call its basic premise into question: as noted by Doron & Rappaport Hovav (2009: 94), reflexives are distinct from transitive verbs with anaphors, so there is nothing to be gained by saying that a reflexive in hitXaY̯ eZ has different morphology than the XiY̯ eZ form and the same semantics. My analysis does face a potential weakness when considering quantificational subjects. Assuming that bound variables are interpreted in their base-generated positions, the semantics of (51) wrongly generates the possibility that each boy shaved another boy, such that in total every boy shaved one other boy.
(51) kol ha-nearim hitgalx-u kol ha-ne'arim. all the-youngsters shaved.intns.mid-3pl ✓ 'All the boys shaved.' The problem of quantifiers does not arise on the silent anaphor approach: the high quantifier simply binds its lower variable. Unfortunately, we have already seen how that approach cannot handle the basic, non-quantified facts. It seems, then, that the details of my semantics require an additional amendment. Perhaps it will be possible to QR the quantifier independently of the restrictor, or merge the quantifier after movement has taken place; I will not develop the idea here.
Another alternative would be to suggest that hitXaY̯ eZ is itself a reflexivizer, as proposed by Reinhart & Siloni (2005), effectively assuming that Hebrew reflexives are unergative. However, I argue in Section 6.2 that such a proposal would give up on any attempt to explain why it is precisely hitXaY̯ eZ (and not any other template) in which reflexive verbs appear: why is the reflexivizer the same form as an anticausativizer, and why the form with a prefix and non-spirantization? Importantly, this template is morphophonologically complex (prefix and de-spirantization) and also semantically complex. The two phenomena were tied together by implicating two functional heads in the structure: marked morphophonology signals internal structure and corresponds to marked syntax and semantics, again the result of complex internal structure.
The differences between the semantic approaches are summarized in (52). On balance, it appears that something special needs to be said about reflexives within the VoiceP domain, but at least for the Hebrew case the current proposal requires fewer stipulations.
(52) Reflexivity in hitXaY̯ eZ: strengths and weaknesses of different frameworks.

Current system Traditional approaches
Mechanism Movement Silent reflexive anaphor or: Arity reduction or: z-Combinator

Advantages
No reflexivizers Standardly assumed Compose independent heads Disadvantages Quantified subjects Stipulated reflexivizers Cross-templatic overgeneration Additional assumptions required

Afto-reflexives in Greek
The analysis of hitXaY̯ eZ reflexives is similar to a recent analysis of reflexive verbs in Greek. In that language, reflexives can be derived by using Voice [-D] and a prefixal modifier afto-; the example elaborated on below is afto-katijori-thike 'accused himself/herself'. In work on this construction, Alexiadou & Schäfer (2013), Alexiadou (2015) and Spathas et al. (2015) argue that these reflexives are the result of combining two morphemes with the root: a Middle Voice head and the bound anti-assistive intensifier afto-. Let us see how their analysis works and then how it compares with ours.

The proposal
Greek Middle Voice is typical of anticausative, passive and middle verbs in the language, similar to Voice [-D] (Alexiadou & Doron 2012). Mediopassives as in (53) are unaccusative. (53) Greek medio-passive a. I Maria katijori-thike. the Maria accused-nact.3sg 'Maria got accused', 'Maria was accused.' b. 30 Kastner In work on this construction, Alexiadou & Schäfer (2013), Alexiadou (2015) and Spathas et al. (2015) argue that these reflexives are the result of combining two morphemes with the root: a Middle Voice head and the bound anti-assistive intensifier afto-. Let us see how their analysis works and then how it compares with ours.

The proposal
Greek Middle Voice is typical of anticausative, passive and middle verbs in the language, similar to Voice [-D] (Alexiadou & Doron 2012). Mediopassives as in (53)  For certain roots, a reflexive can be built on the basis of medio-passives like (53). The reflexive construction is derived using afto-, an "anti-assistive intensifier" similar to √ action and to English non-reflexive herself which is exemplified in (54).
(54) She built the house herself anti-assistive .
According to Spathas et al. (2015), the Greek equivalent of Voice [-D] blocks an external argument from being merged in its specifier, but the internal argument is allowed to undergo A-movement to Spec,VoiceP. The prefix afto-then "tucks in" and adjoins to VoiceP, (55b-c). (55) Greek reflexive For certain roots, a reflexive can be built on the basis of medio-passives like (53). The reflexive construction is derived using afto-, an "anti-assistive intensifier" similar to Hebrew 9 inally proposed as the functional head introducing external arguments . The typology of Voice heads follows recent proposals by Schäfer (2008) uening (2014) for English, Alexiadou & Doron (2012) and Spathas et al. ek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icelandic, Zu (2015) for Latin. What this kind of framework does is to separate the emantic behavior of Voice heads, leading to new analyses of argument ations Schäfer 2017). to be instantiated in three different ways, depending on whether it has a [-D] feature or is underspecified for [±D]. This is an EPP feature, either in the specifier of its head, prohibiting a DP in its specifier or remainout the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to ce head which bans the merger of a DP in its specifier, leaving the struccanonical subject.

ice [-D]
Voice head with a [-D] feature, prohibiting anything with a [D] feature m merging in its specifier. typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] es not assign accusative case itself (Chomsky 1995) or through the callation of dependent case (Marantz 1991 and to English non-reflexive herself which is exemplified in (54).
(54) She built the house herself anti-assistive .
According to Spathas et al. (2015), the Greek equivalent of Voice [-D] blocks an external argument from being merged in its specifier, but the internal argument is allowed to undergo A-movement to Spec,VoiceP. The prefix afto-then "tucks in" and adjoins to VoiceP, (55b-c). (55) Greek reflexive a. I Maria afto-katijori-thike the Maria self-accused-nact.3sg 'Maria accused herself.' b.
Reflexive verbs in Hebrew 31

b. VoiceP
Maria Maria afto VoiceP1 Voice [-D] -thike vP v+ √ accuse katijori DP In the semantics, afto states that its associate (the internal argument, being the only argument in the structure) is the sole Agent in every sub-event of the event, so that Agent and Theme end up being coreferential. The denotations in (56) are adapted from Spathas et al. (2015Spathas et al. ( : 1330Spathas et al. ( , 1332Spathas et al. ( , 1335 In the semantics, afto states that its associate (the internal argument, being the only argument in the structure) is the sole Agent in every sub-event of the event, so that Agent and Theme end up being coreferential. The denotations in (56) are adapted from Spathas et al. (2015Spathas et al. ( : 1330Spathas et al. ( , 1332Spathas et al. ( , 1335 [A] description of events of someone accusing Mary such that Mary is the agent in all sub-events of that event. This is a reflexive interpretation." The two affixes -thike (Voice [-D] ) and afto-(anti-assistive ≈ √ action) combine to give a reflexive reading: the former contributes by restricting the number of arguments In the semantics, afto states that its associate (the internal argument, being the only argument in the structure) is the sole Agent in every sub-event of the event, so that Agent and Theme end up being coreferential. The denotations in (56) are adapted from Spathas et al. (2015Spathas et al. ( : 1330Spathas et al. ( , 1332Spathas et al. ( , 1335 "[A] description of events of someone accusing Mary such that Mary is the agent in all sub-events of that event. This is a reflexive interpretation." The two affixes -thike (Voice [-D] ) and afto-(anti-assistive ≈ Reflexive verbs in Hebrew ner 2016), originally proposed as the functional head introducing external argum (Kratzer 1996). The typology of Voice heads follows recent proposals by Schäfer (2 for German, Bruening (2014) for English, Alexiadou & Doron (2012) and Spathas (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for Icela and Kastner & Zu (2015) for Latin. What this kind of framework does is to separa syntactic and semantic behavior of Voice heads, leading to new analyses of argu structure alternations Schäfer 2017). I take Voice to be instantiated in three different ways, depending on whether it [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, e requiring a DP in the specifier of its head, prohibiting a DP in its specifier or re ing agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly re Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the ture without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] fe from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Vo does not assign accusative case itself (Chomsky 1995) or through th culation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in a guages. In Hebrew, it has been argued that the three variants have distinct expo (Kastner 2016; ) combine to give a reflexive reading: the former contributes by restricting the number of arguments in the clause and the latter contributes by specifying that the existing argument is the only Agent. To reiterate, there is no dedicated reflexivizer in this structure.
This kind of derivation also depends on the lexical semantics of the root. Afto-reflexives are not possible with the Greek class of Naturally Reflexive roots (similar to Self-Oriented roots, Spathas et al. 2015Spathas et al. : 1337. It is only in combination with afto-and the right root class that the internal and external arguments are forced to be coreferential, leading to a reflexive interpretation.

Heads and roots crosslinguistically
A number of differences between the analyses of Greek and Hebrew are highlighted in (57). In all cases, the differences are expressed in terms that are natural in the current framework.
Greek afto-attaches to the extended projection of Voice [-D] , whereas Hebrew quiring a DP in the specifier of its head, prohibiting a DP in its specifier or remaing agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to ice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the strucre without a canonical subject. 9) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all lanages. In Hebrew, it has been argued that the three variants have distinct exponents astner 2016; 2017). 0) a.
Voice [-D] ↔ niXYaZ c. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two additional templates which e have already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ and tXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I explain how brings about an external argument without requiring one in the syntax (which is what e feature [+D] does, but Voice [+D] will not be discussed further.).
tational matters: the template heXYiZ usually appears in the literature as hiXYiZ, with the first vowel /i/. Speakers of my generation and at least one generation older use the /e/ form, and so I transcribe /e/ roughout. Conversely, the initial /h/ is usually dropped in speech. I retain it for two reasons. First, /h/ is ll pronounced by some older speakers and certain sociolinguistic groups, especially marginalized ones chwarzwald 1981; Gafter 2014b). Second, the initial h-should help the non-Semitist reader distinguish is template from others. appears that √ action does not combine with Voice [+D] ; see Kastner (2016: Chapter 2.4.1) for discusn.
modifies either Voice or Voice [-D] . b. Greek afto-is incompatible with Naturally Reflexive roots, whereas Hebrew I take Voice to be instantiated in three different ways, depending on whether it has a D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP feature, either quiring a DP in the specifier of its head, prohibiting a DP in its specifier or remaing agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly refer to ice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving the strucre without a canonical subject. 9) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D] feature from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice, Voice [-D] does not assign accusative case itself (Chomsky 1995) or through the calculation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated in all lanages. In Hebrew, it has been argued that the three variants have distinct exponents astner 2016; 2017). 0) a.
Voice [-D] ↔ niXYaZ c. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two additional templates which e have already seen in (4)-(6) and which lie at the heart of the current study, XiY̯ eZ and tXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I explain how brings about an external argument without requiring one in the syntax (which is what e feature [+D] does, but Voice [+D] will not be discussed further.).
tational matters: the template heXYiZ usually appears in the literature as hiXYiZ, with the first vowel /i/. Speakers of my generation and at least one generation older use the /e/ form, and so I transcribe /e/ roughout. Conversely, the initial /h/ is usually dropped in speech. I retain it for two reasons. First, /h/ is ll pronounced by some older speakers and certain sociolinguistic groups, especially marginalized ones chwarzwald 1981; Gafter 2014b). Second, the initial h-should help the non-Semitist reader distinguish is template from others. appears that √ action does not combine with Voice [+D] ; see Kastner (2016: Chapter 2.4.1) for discusn.
combines with all classes of roots. c. Greek afto-uses counter-cyclic attachment, whereas Hebrew structure alternations Schäfer 2017).
I take Voice to be instantiated in three different ways, depending [+D] feature, a [-D] feature or is underspecified for [±D]. This is an requiring a DP in the specifier of its head, prohibiting a DP in its ing agnostic about the specifier. In the discussion of hitXaY̯ eZ we Voice [-D] , a Voice head which bans the merger of a DP in its specifi ture without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anythin from merging in its specifier. As typically assumed for unaccusative little v or unaccus does not assign accusative case itself (Chomsky 1995) culation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily in guages. In Hebrew, it has been argued that the three variants have (Kastner 2016;. (20) a.
Voice [-D] ↔ niXYaZ c. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two addition we have already seen in (4)-(6) and which lie at the heart of the curr hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, w it brings about an external argument without requiring one in the sy the feature [+D] does, but Voice [+D] will not be discussed further.).
3 Notational matters: the template heXYiZ usually appears in the literature as hiX an /i/. Speakers of my generation and at least one generation older use the /e/ form throughout. Conversely, the initial /h/ is usually dropped in speech. I retain it for still pronounced by some older speakers and certain sociolinguistic groups, espec (Schwarzwald 1981;Gafter 2014b). Second, the initial h-should help the non-Sem this template from others. 4 It appears that √ action does not combine with Voice [+D] ; see Kastner (2016: C sion. is satisfied through delayed saturation. The first two differences can only be stated in the kind of vocabulary developed in the current framework. This enterprise-of which Spathas et al. (2015) is part-allows us to pinpoint similarities and differences between the languages by analyzing functional heads and their interaction with roots. If this approach is on the right track, work on argument structure will be in a better position to understand how syntactic structure feeds the semantics and how these combinations are reflected in the morphology.
The last difference, (57c), is mostly a matter of technical preference. The proposal for Greek assumes that nothing can be merged in the specifier of Voice [-D] but that the internal argument may be moved there. I prefer to maintain a general "Merge over Move" principle of economy, which is inconsistent with such an analysis unless further assumptions about the numeration are made explicit. Similarly, I do not adopt late merger of adverbials such as afto. That said, the bottom line of the current analysis-the functional heads and the way in which roots constrain semantic interpretation-can be implemented in the minimally different afto-style system as well.
In Greek, like in Hebrew, the case can be made for a distinction between Self-Oriented and Other-Oriented roots. This approach to roots can be extended further. In her work on Italian verbs of motion, Folli (2001) describes different kinds of motion events which lead to verbs that are either unaccusative, unergative or ambiguous between the two. These events correspond to different root classes. Levinson (2014) likewise shows that verbs of creation in English license different kinds of syntactic constructions depending on their own semantics. For a summary of additional cases see Alexiadou et al. (2014). Furthermore, it has been shown that even within an existing change-of-state alternation, the transitive verb may have a meaning that the intransitive does not (Levin & Rappaport Hovav 1995: 85): (58) a. He broke his promise/the contract/the world record. b. *His promise/The contract/The world record broke. There is a clear sense in which the lexical semantics of different roots dictates the kind of derivations they may participate in. These observations all call for further research and formalization.
6 Alternatives for Hebrew

Distributed morphosemantics (Doron 2003)
Like the current system, the seminal analysis of Hebrew verbs in Doron (2003) employed a number of functional heads to derive the different templates. Doron (2003) was the first to identify basic non-templatic elements that combine compositionally in order to form Hebrew verbs. For example, a MIDDLE head μ was used to derive the "middle" template niXYaZ, where I make use of the head Voice [-D] familiar from other languages.
The present system is influenced directly by Doron's. The important conceptual difference is that my elements are syntactic whereas those in Doron (2003) can be characterized as morphosemantic: each one had a distinct semantic role. A Doron-style system takes the semantics as its starting point, attempting to reach the templates from syntactic-semantic primitives signified by the functional heads. Such a system runs into the basic problem of Semitic morphology: one cannot map the phonology directly onto the semantics. For example, there is no way in which a causative verb has a unique morphophonological exponent.
Reflexive verbs highlight a false prediction made by this system. Doron (2003: 60) derives reflexives in hitXaY̯ eZ by assuming that a head MIDDLE assigns the Agent role for this root. This explains why histager 'secluded himself' is agentive, hence reflexive. However, if the only relevant elements are Voice [-D] and the root, then a verb in the same root in niXYaZ (where I have Voice [-D] and Doron 2003 has MIDDLE) is also predicted to be agentive. This expectation is incorrect: nisgar 'closed' is unaccusative. That analysis is almost a mirror image of the one presented here: while I let Reflexive verbs in Hebrew ner 2016), originally proposed as the functional head introducing external ar (Kratzer 1996). The typology of Voice heads follows recent proposals by Schäfe for German, Bruening (2014) for English, Alexiadou & Doron (2012) and Spat (2015) for Greek, Alexiadou & Doron (2012) for Hebrew, Wood (2015) for I and Kastner & Zu (2015) for Latin. What this kind of framework does is to sep syntactic and semantic behavior of Voice heads, leading to new analyses of a structure alternations Schäfer 2017).
I take Voice to be instantiated in three different ways, depending on whethe [+D] feature, a [-D] feature or is underspecified for [±D]. This is an EPP featu requiring a DP in the specifier of its head, prohibiting a DP in its specifier or ing agnostic about the specifier. In the discussion of hitXaY̯ eZ we will mostly Voice [-D] , a Voice head which bans the merger of a DP in its specifier, leaving t ture without a canonical subject. (19) a.

Voice [-D]
A Voice head with a [-D] feature, prohibiting anything with a [D from merging in its specifier. As typically assumed for unaccusative little v or unaccusative Voice does not assign accusative case itself (Chomsky 1995) or through culation of dependent case (Marantz 1991). b.
Voice [-D] = λP <s,t> .P The tripartite classification of Voice heads is not necessarily instantiated i guages. In Hebrew, it has been argued that the three variants have distinct ex (Kastner 2016;. (20) a.
Voice [-D] ↔ niXYaZ c. Voice [+D] ↔ heXYiZ 3 Adding an agentive modifier, √ action, derives the two additional templat we have already seen in (4)-(6) and which lie at the heart of the current study, X hitXaY̯ eZ. 4 This element is discussed in more detail in Section 3.2, where I exp it brings about an external argument without requiring one in the syntax (whic the feature [+D] does, but Voice [+D] will not be discussed further.).
3 Notational matters: the template heXYiZ usually appears in the literature as hiXYiZ, with the an /i/. Speakers of my generation and at least one generation older use the /e/ form, and so I tra throughout. Conversely, the initial /h/ is usually dropped in speech. I retain it for two reasons. still pronounced by some older speakers and certain sociolinguistic groups, especially margina (Schwarzwald 1981;Gafter 2014b add agentivity to a structure with Voice [-D] , thereby deriving reflexives, the morphosemantic account invokes added agentivity for certain roots, bypassing the syntax in ways that lead to false predictions. While this problem can be overcome, the system as a whole has little to say about the unaccusative (for anticausatives) and unergative (for reflexives) characteristics of verbs in hitXaY̯ eZ, since it is not based strictly in the syntax.
I should take a moment to emphasize the most important gains of the morphosemantic theory. Treating templates as emergent from heads that do separate syntactic and semantic work gave us a new way to analyze argument structure alternations across templates, based on a wealth of empirical data. The theory also made a compelling case for the root as an atomic element participating in the derivation, making a number of novel observations along the way. Where we have made progress is by flipping one of the assumptions on its head: that the primitives have strict syntactic content and flexible semantic content, rather than strict semantic content and unclear syntactic content.

Templates as morphemes
In juxtaposition to an "emergent" view of templates from functional heads, the traditional approach to Semitic templates has been to treat them as independent atomic elements, i.e. morphemes. Contemporary work in this vein spans highly divergent implementations but includes Arad (2003Arad ( , 2005, who treated verbal templates as distinct spell-outs of Voice; Borer (2013), for whom different templates are different "functors"; Aronoff (1994Aronoff ( , 2007, who identifies templates with conjugation classes; and Reinhart & Siloni (2005) and Laks (2011Laks ( , 2014, whose lexicalist account similarly grants morphemic status to verbal templates. Syntactic and lexicalist accounts both need to stipulate that only a subset of roots (or stems) licenses reflexive derivations. What is at issue here is the status of the template. The general problem with morphemic approaches to templates is that a given template simply does not have a deterministic syntax or semantics, as noted in the Introduction and argued for by Doron (2003) and Kastner (2016). Arad (2005: 198) and Borer (2013: 564) actually speculate that a configurational approach (like in our theory) might be more viable than a feature-based or functor-based approach. As far as the treatment of reflexives is concerned, morphemic accounts can go no further than stipulating that hitXaY̯ eZ is the template for reflexive verbs.
To repeat a point made earlier: stipulating that reflexives are formed using the morphophonological form hitXaY̯ eZ does not explain why it is precisely this template that is involved, nor why this template also allows for anticausativization. Certain correlations would then be missed out on: that this template is both morphophonologically and semantically complex, or that reflexives and anticausatives appear to have a shared base. The system developed in this paper provided the answer to this question, based on functional heads required elsewhere in the grammar.

Conclusion
The main empirical issue addressed in this paper was the morphology of reflexive verbs in Hebrew. On the morphophonological side, these verbs have a specific form, though this form is shared with other kinds of verbs (anticausatives and reciprocals). On the morphosyntactic side, reflexive verbs show mixed behavior of unergative and unaccusative constructions. And on the morphosemantic side, they seem to only be available with a certain class of lexical roots. The resulting theoretical discussion addressed the question of how semantic roles are distributed in the syntax and how are they reflected in the morphology.
I have argued for a system in which neither theta-roles nor dedicated reflexivizers are necessary. Instead, a nonactive syntactic structure-that is, one without an external argument-can have agentive semantics if and only if it is coupled with an agentive modifier and an appropriate root. In other words, the correct interpretation of the verb is a result of functional heads combining with the idiosyncratic information contained in the root.
The analysis presented here is part of a general approach to non-concatenative morphology of the Semitic kind as exemplified by Hebrew. Generally, one cannot predict the meaning of a verb from its morphophonological form (its template), nor can one predict what template a verb will have based solely on its meaning. The solution to this mapping problem was implemented in a system that builds syntactic structure and then interprets said structure at PF and LF. For a consistent system to be set up, templates must be viewed as emergent from functional heads in the structure and not as morphemes, which is the traditional view. Once the structure is set up correctly, roots have the power to influence the interpretation at the semantics, but no earlier. How they do this is a matter of ongoing work, in Semitic and beyond.