45 LANGUAGE
The Middle Construction in Minimalist Syntax: 
English and Slovenian
ABStRA ct
The paper discusses the argument structure of the English middle construction and its Slovenian 
equivalent from the perspective of minimalist syntax. The paper first introduces Bruening’s 
(2012) recent approach to syntactic middle formation, which posits that middle sentences are 
derived via an operator that existentially quantifies over the open agent variable introduced by an 
active v oice projection. Subsequently, the paper argues that the adverbial modifier in the middle 
construction is not a semantic argument of the null operator, contra Bruening (2012). Finally, 
the paper proposes that the reflexive morpheme se in the related Slovenian se-sentences plays a 
role of valency reduction similar to that of the null English operator.
Keywords: valency; middle construction; reflexive morpheme; syntactic minimalism; argument 
structure 
Srednjiška zgradba v minimalistični skladnji: 
angleščina in slovenščina
POvZEtEK
Pričujoči prispevek obravnava vezljivost angleške in slovenske srednjiške zgradbe z vidika 
minimalistične skladnje. Prispevek najprej predstavi Brueningovo (2012) skladenjsko 
razčlembo srednjiške zgradbe, kjer ključni del skladenjske izpeljave sloni na fonološko ničtem 
skladenjskopomenskem operatorju. v formalnopomenskem smislu je tovrstni operator funkcija, 
ki uvaja eksistencialni kvantifikator, ta pa se v pomenski ravnini veže s spremenljivko, ki jo 
uvaja jedro zveze glagolskega načina voice. Prispevek tudi predloži, v nasprotju z Brueningom 
(2012), da prislovna zveza v srednjiški zgradbi ni pomenski argument ničtega operatorja. v 
povezavi s slovenščino prispevek argumentira, da prosti morfem se igra podobno vlogo v povezavi 
z vezljivostjo kot angleški ničti operator.
Ključne besede: vezljivost; srednjiška zgradba; povratni morfem; minimalistična skladnja; 
argumentna struktura
Jakob Lenardič
University of Ljubljana, Slovenia
2018, v ol. 15 (2), 45–61(120)
revije.ff.uni-lj.si/elope
doi: 10.4312/elope.15.2.45-61
UDc: 811.111’367:811.163.6

46 Jakob Lenardič  The Middle Construction in Minimalist Syntax: English and Slovenian
The Middle Construction in Minimalist Syntax: 
English and Slovenian
1 Introduction
The argument structure of the middle construction (1) has been a long-standing grammatical puzzle. 
(1) Politicians bribe easily.
The main point of contention in the formal literature relates to the syntactic representation 
of the implied agent argument of (1); that is, the person or people who can easily bribe the 
politicians. The problem is that the construction behaves inconsistently with respect to tests that 
diagnose the structural presence of the agent. On the one hand, the middle sentence in (1) differs 
from a prototypical passive construction (2) in that it cannot licence the by- phrase (3), by means 
of which the implied agent can be represented overtly. 
(2) The politician was bribed (by John).
(3) Politicians bribe easily (*by John).
What is puzzling, however, is that middles are able to select an instrumental with-phrase (4), 
which is also allowed in the passive (5), but not in the ergative anticausative construction (6), 
which is standardly analysed as completely omitting the agent argument in its syntactic, as well as 
semantic, representation (for an overview of recent approaches, see Alexiadou Anagnostopoulou, 
and Schäfer 2015, chapter 3). In this sense, it is often observed that the agent in the middle is 
only “implicit” (Klingvall 2005), so absent from the syntactic derivation.
(4) The door opens easily with a key.
(5) The door was opened with a key.
(6) The door opened (*with a key). 
The cross-linguistic grammatical properties of middles differ widely from language to language 
(Lekakou 2004). For instance, the Slovenian middle equivalent of sentence (4), which is given 
in (7), contains in its structure an additional category in the form of the reflexive morpheme se.
(7)  Vrata            se         zlahka odprejo s     ključem.
doors.nom   refl       easily    open      with key
“The door opens easily with a key.”
Apart from se, Slovenian middles differ from English ones in that they do not involve the so-
called generic repair strategy (Härtl 2012), which means that the majority of predicates are licit 
in the English middle construction if and only if the interpretation is generic (8b), whereas the 
equivalent of eventive (8a) is perfectly fine in Slovenian (9a).
(8)  a. *The bread cut.                                        
     Intended: “Somebody cut the bread.” 
b.  The bread cuts easily.
                                             (Levin and Rappaport-Hovav 1995, 95)

47 LANGUAGE
(9)  a. Kruh              se      je        rezal.
    bread.nom     refl   aux       cut
    “Somebody was cutting the bread.”
b. Kruh              se       zlahka reže.
    bread.nom    refl    easily   cut
    “The bread cuts easily.”
The aim of this paper is to discuss how middle sentences can be syntactically accounted for, taking 
into consideration both their selection properties stated in (2–6) with respect to the implicit agent 
and the cross-linguistic structural properties that distinguish English middles from Slovenian ones 
(7–9). In terms of theory, the paper will follow a syntactic approach in the spirit of constructional 
minimalism (Ramchand 2008, Bruening 2012, Alexiadou, Anagnostopoulou, and Schäfer 2015, 
to name a few), in which the formation of lexically complex predicates, such as those in the middle 
construction, takes place entirely during the syntactic derivation. Within this framework, the 
paper will first present and discuss the empirical and theoretical advantages of Bruening’s (2012) 
recent proposal of syntactic middle formation in English, where the construction is derived by 
means of a null operator that targets the open agent variable that is introduced by a voice head. 
Subsequently, the paper proposes that the adverb, although obligatory in many middles, is not a 
semantic requirement of the null operator, contra Bruening (2012). Finally, the paper will discuss 
how Slovenian se-sentences are derived, and propose that se plays a similar role in semantics to the 
English middle operator, but has fewer selectional requirements. In this respect, the paper will also 
propose that sentences (9a) and (9b) are structurally the same (i.e., se has the same function in both 
sentences, in that it binds the open agent variable introduced by v oice), with the only difference 
being that the event variable in (9b) is bound by a generic quantifier. 
The paper is organized as follows. Section 2 lays out the theoretical framework and presents 
Bruening’s (2012) proposal as an alternative to the traditional lexical account of middle formation, 
and discusses how his syntactic approach elegantly accounts for the key properties of English 
middles, such as the arbitrary reference of the agent, the generic interpretation, and the licensing 
of the with-phrase. Section 3 discusses the role of the adverb in the middle construction in relation 
to Bruening’s (2012) operator. Section 4 discusses related Slovenian se-sentences and proposes a 
semantics for se contrasted with the English null middle operator. Section 5 is the conclusion.
2 Deriving Middles – From the Lexical Approach to a 
Minimalist Syntactic Account
2.1 Middle Formation in a Lexical Framework
The traditional approach to the derivation of the middle construction is the lexical account, in 
which the middle construction (11) is understood as a pre-syntactic derivational transformation 
(i.e., detransitivisation) of an otherwise active transitive sentence (10). 
(10) John reads this book easily.
(11) This book reads easily.

48 Jakob Lenardič  The Middle Construction in Minimalist Syntax: English and Slovenian
Ackema and Schoorlemmer (1994, 65–67) describe a very detailed implementation of such an 
idea, which – at least to some degree – informs much subsequent work on middles (e.g., Marelj 
2004 for Slavic). The gist of their proposal is that there is a pre-syntactic level of representation 
called Lexical Conceptual Structure, which contains lexical rules that can modify the argument 
structure of a predicate like read before syntactic projection takes place. In this sense, they propose 
that there is a rule called Middle Formation that pre-syntactically transforms an otherwise active 
transitive entry, (12), into a middle one, (13).
(12) read
active
 (θ
Actor
, (θ
patient
))
ext
(13) read
middle
 (ARB, (θ
patient
))
ext
Such a lexical approach is conceptually very appealing, as it accounts for a host of disparate 
grammatical facts of the middle construction. For instance, the transformation of the lexical entry 
read from (12) to (13) shows that the Middle Formation is an operation in which the agent θ-role 
is pre-syntactically realised as a non-projecting arbitrary participant (ARB in (13)). Since the agent 
is thereby saturated in the lexicon, it is not present in the syntactic component, which explains 
why middles cannot license the by -phrase, whose admissibility is otherwise assumed by the lexical 
approach to be contingent on some kind of structural presence of the external argument (e.g., for a 
passive sentence, which admits the by- phrase in contrast with a middle, Marelj (2004) assumes that 
the external argument is present in the syntax as a demoted oblique argument).
Notice, lastly, that lexical middle formation posits an argument-linking rule called Externalize, 
abbreviated with the superscript Ext in (12) and (13). This rule entails that the remaining θ-role 
is realised vP-externally; that is, as an external argument. The rule is assumed in order to explain 
a host of cross-linguistic data, such as the fact that middles morphosyntactically pattern with the 
intransitive unergative construction, which selects an external argument. For instance, in Dutch, 
middles – like unergatives – select the auxiliary verb have (14) in periphrastic tense constructions, 
whereas unaccusatives, which realise their sole argument vP-internally, select be (15).
(14) Dit vlees heeft/*is altijd gemakkelijk gesneden. (Dutch middle)
    this meat has   /  is always easily             cut.
    “This meat has always cut well.”
(15) De spanning is/*heeft dagenland toegenomen de spanning.   (Dutch unaccusative)
    the  tension   is/ has      for-days    increased.
    “The tension has increased for days.”
 (Examples taken from Ackema and Schoorlemmer 1994, 61, 64)
2.2 Introducing the External Argument in a Minimalist Theory 
of Grammar
In contemporary theories of grammar, however, there has been a longstanding aim to reduce the 
complexity of such a model – i.e., to make the grammar “more minimalist” in chomsky’s terms 
(2001) – by placing all the compositional burden of word-formational processes on the syntactic 
component (e.g., Marantz 1997; Ramchand 2008). One of the problems of a non-minimalist 
grammar that employs pre-syntactic rules for altering argument structure is that it invariably 

49 LANGUAGE
leads to what Ramchand (2008, 15) calls “massively stipulated homonymies”; in the case of 
middles, this is the idea that there are actually two separate entries for the verb read – one regular 
active (12) and one middle (13) – in the English lexicon, which is additionally suspect from 
the point of view of learnability (Borer 2004).
1
 consequently, what is needed from the point 
of view of such constructional minimalism is a theory of middles in which the structural and 
selectional quirks of the construction (i.e., the licensing of with-phrases and the inadmissibility 
of by-phrases, their generic interpretation, their patterning with unergatives) are all handled 
during the syntactic derivation.
I start the discussion by looking at how the argument structure of a predicate is built in a 
minimalist theory of language. Since the main concern is the status of the agent in middles, 
the crucial component for this purpose is Kratzer’s (1996) influential proposal that the external 
thematic role of a predicate enters the syntactic derivation outside the lexical v erb Phrase, which 
hosts solely the internal argument. In her proposal, a verb like read lexically denotes a relation 
between an event of reading and an entity (i.e., the patient) that is read in the event (16). 
(16) ⟦read⟧ = λx
⟨e⟩
.λe
⟨s⟩
.[read(e) ∧ patient(x, e)]
Note that the way the syntactic structure is built closely reflects the semantic requirements of 
a given predicate. concretely, the merger of the lexical item read and the DP the book, which 
results in the syntactic structure in (17), semantically corresponds to functional application, in 
that the verb read denotes a function (16) that selects the denotation of the DP the book as its 
first semantic argument, as shown in (18).
(17) The syntax of VP 
  
structural and selectional quirks of the construction (i.e., the licensing of with-phrases and the 
inadmissibility of by-phrases, their generic interpretation, their patterning with unergatives) are
all handled during the syntactic derivation.
I start the discussion by looking at how the argument structure of a predicate is built in a 
minimalist theory of language. Since the main concern is the status of the agent in middles, the 
crucial component for this purpose is Kratzer’s (1996) influential proposal that the external 
thematic role of a predicate enters the syntactic derivation outside the lexical Verb Phrase, 
which hosts solely the internal argument. In her proposal, a verb like read lexically denotes a
relation between an event of reading and an entity (i.e., the patient) that is read in the event 
(16).
(16) ⟦read ⟧ = λx ⟨e ⟩.λe ⟨s ⟩.[READ(e) ∧ PATIENT(x, e)]
Note that the way the syntactic structure is built closely reflects the semantic requirements of 
a given predicate. Concretely, the merger of the lexical item read and the DP the book, which 
results in the syntactic structure in (17), semantically corresponds to functional application, in
that the verb read denotes a function (16) that selects the denotation of the DP the book as its 
first semantic argument, as shown in (18).
(17) The syntax of VP
(18) The semantics of merging V and DP
a. ⟦VP ⟧ = ⟦V ⟧( ⟦DP ⟧) = ⟦read ⟧( ⟦the book ⟧) = 
b. λx ⟨e ⟩.λe ⟨s ⟩.[READ(e) ∧ PATIENT(x, e)](the book) = 
c. λe ⟨s ⟩.[READ(e) ∧ PATIENT(the book, e)]
At the next step of structure building, the external participant is introduced into the semantics 
of VP. To this end, VP merges in Kratzer’s system (1996) with a functional head labelled 
Voice, as in (19). Semantically, this head expands the denotation of VP, given in (18c), by 
transforming it into a two-argument function that selects an additional participant at the level 
of the intermediate projection Voice’; since read lexically denotes a dynamic event, the Voice
head identifies this participant as the agent of the event.
2
(19) The syntax and semantics of VoiceP without an external argument
                                                
2
Some of the examples in the paper have bare plural nouns as internal arguments (e.g., Physics books read easily).
Strictly speaking, such DPs are semantically complex functions, where they combine via Functional Application with the predicate in the same manner 
as the simple definite descriptor, which is of type ⟨e ⟩ in (18c). 
(18) The semantics of merging V and DP
  
structural and selectional quirks of the construction (i.e., the licensing of with-phrases and the 
inadmissibility of by-phrases, their generic interpretation, their patterning with unergatives) are
all handled during the syntactic derivation.
I start the discussion by looking at how the argument structure of a predicate is built in a 
minimalist theory of language. Since the main concern is the status of the agent in middles, the 
crucial component for this purpose is Kratzer’s (1996) influential proposal that the external 
thematic role of a predicate enters the syntactic derivation outside the lexical Verb Phrase, 
which hosts solely the internal argument. In her proposal, a verb like read lexically denotes a
relation between an event of reading and an entity (i.e., the patient) that is read in the event 
(16).
(16) ⟦read ⟧ = λx ⟨e ⟩.λe ⟨s ⟩.[READ(e) ∧ PATIENT(x, e)]
Note that the way the syntactic structure is built closely reflects the semantic requirements of 
a given predicate. Concretely, the merger of the lexical item read and the DP the book, which 
results in the syntactic structure in (17), semantically corresponds to functional application, in
that the verb read denotes a function (16) that selects the denotation of the DP the book as its 
first semantic argument, as shown in (18).
(17) The syntax of VP
(18) The semantics of merging V and DP
a. ⟦VP ⟧ = ⟦V ⟧( ⟦DP ⟧) = ⟦read ⟧( ⟦the book ⟧) = 
b. λx ⟨e⟩.λe ⟨s ⟩.[READ(e) ∧ PATIENT(x, e)](the book) = 
c. λe ⟨s ⟩.[READ(e) ∧ PATIENT(the book, e)]
At the next step of structure building, the external participant is introduced into the semantics 
of VP. To this end, VP merges in Kratzer’s system (1996) with a functional head labelled 
Voice, as in (19). Semantically, this head expands the denotation of VP, given in (18c), by 
transforming it into a two-argument function that selects an additional participant at the level 
of the intermediate projection Voice’; since read lexically denotes a dynamic event, the Voice
head identifies this participant as the agent of the event.
2
(19) The syntax and semantics of VoiceP without an external argument
                                                
2
Some of the examples in the paper have bare plural nouns as internal arguments (e.g., Physics books read easily).
Strictly speaking, such DPs are semantically complex functions, where they combine via Functional Application with the predicate in the same manner 
as the simple definite descriptor, which is of type ⟨e ⟩ in (18c). 
At the next step of structure building, the external participant is introduced into the semantics of 
vP . to this end, vP merges in Kratzer’s system (1996) with a functional head labelled v oice, as in 
(19). Semantically, this head expands the denotation of vP , given in (18c), by transforming it into 
a two-argument function that selects an additional participant at the level of the intermediate 
projection v oice’; since read lexically denotes a dynamic event, the voice head identifies this 
participant as the agent of the event.
2
1
 Such stipulated homonymy is even more apparent with causative predicates, where it is assumed in the lexicalist framework 
that there are three separate entries for the verb open – (i) the anticausative entry, as in This door opened
anticausative
, created by the 
process of anticausativisation, which involves pre-syntactic external argument deletion (Levin and Rappaport Hovav 1995); 
(ii) the middle entry, as in This door opens
middle
 easily, created by Middle Formation as described in Section 2.1; and (iii) the 
default causative active entry, as in John opened
causative
 the door.
2
 Some of the examples in the paper have bare plural nouns as internal arguments (e.g., Physics books read easily). Strictly 
speaking, such DPs are semantically complex functions, but for the purposes of this paper, I assume a simplified representation 

50 Jakob Lenardič  The Middle Construction in Minimalist Syntax: English and Slovenian
(19) The syntax and semantics of V oiceP without an external argument
                    
To derive a prototypical transitive sentence such as John read the book, what the syntactic 
computation needs to do is simply merge the DP John in the [Spec,VoiceP] position to saturate 
the open variable of Voice’ (20b). The final semantics for VoiceP is given in (20c).
(20) The semantics of VoiceP with external argument John
a. ⟦VoiceP ⟧= ⟦Voice’ ⟧( ⟦John ⟧) = 
b. λx ⟨e ⟩ .λe ⟨s ⟩.[AGENT(x, e) ∧ READ(e) ∧ PATIENT(the book, e)](John) = 
c. λe ⟨s ⟩.[AGENT(John, e) ∧ READ(e) ∧ PATIENT(the book, e)]
2.3 A Syntactic Account of English Middles
There is an agreement in the syntactic literature (e.g., Bruening 2012, Legate 2014, Alexiadou,
Anagnostopoulou, and Schäfer 2015) that VP-external valency reduction (that is, the omission
of an overt external argument in middles, anticausatives, passives, etc.) occurs at the stage in 
which VoiceP is built. The proposal that I will now adopt for middles is taken from Bruening
(2012, 29–31).
Bruening suggests that the derivation of a middle sentence proceeds initially in the same
manner as that of a prototypical active transitive structure, in the sense that a VP containing 
the internal argument first merges with a Voice head introducing the open agent variable. The 
point of syntactic departure in the middle sentence, however, happens because it merges a
semantic operator that selects an unsaturated Voice projection (i.e., the denotation of the
intermediate category Voice’ in (19)), and thereby blocks the merger of a referring DP as the 
external argument. The semantics for such an operator are given in (21).
(21) ⟦MiddleOp ⟧ = λf ⟨e, ⟨s,t⟩⟩ .λ g ⟨s,t ⟩.GNe ∃x[f(x, e) → g(e)]
(Bruening 2012, 31)
According to (21), MiddleOp selects as its first semantic argument a syntactic category that 
logically corresponds to a function with an open agent variable, and maps it onto a scope where
the agent variable is bound by an existential quantifier. The second argument fed into (21) is 
an event modifier of type ⟨s,t ⟩, which corresponds to the adverbial modification in middles 
(i.e., an adverb like easily is interpreted as a function from easy events to truth values; see the 
syntactic tree in (22) below).
Syntactically, note that Bruening’s (2012) proposal is cast within a theory of bare-phrase 
structure where operators can be merged before a grammatical head – Voice, in this case – fully
projects. Differently put, the semantics in (21) suggests that MiddleOp is merged in the [Spec,
VoiceP] position, but this position is canonically (e.g., in X-bar theory, whose notation the
paper has partially adopted) associated with nominal expressions (DPs saturating the agent
variable), and not with operators that select their own semantic arguments. Although such an
assumption is empirically and theoretically well motivated (see the discussion in Bruening 
to derive a prototypical transitive sentence such as John read the book, what the syntactic 
computation needs to do is simply merge the DP John in the [Spec,v oiceP] position to saturate 
the open variable of v oice’ (20b). The final semantics for v oiceP is given in (20c).
(20) The semantics of VoiceP with external argument John
  
To derive a prototypical transitive sentence such as John read the book, what the syntactic 
computation needs to do is simply merge the DP John in the [Spec,VoiceP] position to saturate 
the open variable of Voice’ (20b). The final semantics for VoiceP is given in (20c).
(20) The semantics of VoiceP with external argument John
a. ⟦VoiceP ⟧= ⟦Voice’ ⟧( ⟦John ⟧) = 
b. λx ⟨e⟩ .λe ⟨s ⟩.[AGENT(x, e) ∧ READ(e) ∧ PATIENT(the book, e)](John) = 
c. λe ⟨s ⟩.[AGENT(John, e) ∧ READ(e) ∧ PATIENT(the book, e)]
2.3 A Syntactic Account of English Middles
There is an agreement in the syntactic literature (e.g., Bruening 2012, Legate 2014, Alexiadou,
Anagnostopoulou, and Schäfer 2015) that VP-external valency reduction (that is, the omission
of an overt external argument in middles, anticausatives, passives, etc.) occurs at the stage in 
which VoiceP is built. The proposal that I will now adopt for middles is taken from Bruening
(2012, 29–31).
Bruening suggests that the derivation of a middle sentence proceeds initially in the same
manner as that of a prototypical active transitive structure, in the sense that a VP containing 
the internal argument first merges with a Voice head introducing the open agent variable. The 
point of syntactic departure in the middle sentence, however, happens because it merges a
semantic operator that selects an unsaturated Voice projection (i.e., the denotation of the
intermediate category Voice’ in (19)), and thereby blocks the merger of a referring DP as the 
external argument. The semantics for such an operator are given in (21).
(21) ⟦MiddleOp ⟧ = λf ⟨e, ⟨s,t⟩⟩ .λ g ⟨s,t ⟩.GNe ∃x[f(x, e) → g(e)]
(Bruening 2012, 31)
According to (21), MiddleOp selects as its first semantic argument a syntactic category that 
logically corresponds to a function with an open agent variable, and maps it onto a scope where
the agent variable is bound by an existential quantifier. The second argument fed into (21) is 
an event modifier of type ⟨s,t ⟩, which corresponds to the adverbial modification in middles 
(i.e., an adverb like easily is interpreted as a function from easy events to truth values; see the 
syntactic tree in (22) below).
Syntactically, note that Bruening’s (2012) proposal is cast within a theory of bare-phrase 
structure where operators can be merged before a grammatical head – Voice, in this case – fully
projects. Differently put, the semantics in (21) suggests that MiddleOp is merged in the [Spec,
VoiceP] position, but this position is canonically (e.g., in X-bar theory, whose notation the
paper has partially adopted) associated with nominal expressions (DPs saturating the agent
variable), and not with operators that select their own semantic arguments. Although such an
assumption is empirically and theoretically well motivated (see the discussion in Bruening 
2.3 A Syntactic Account of English Middles
There is an agreement in the syntactic literature (e.g., Bruening 2012, Legate 2014, Alexiadou, 
Anagnostopoulou, and Schäfer 2015) that vP-external valency reduction (that is, the omission 
of an overt external argument in middles, anticausatives, passives, etc.) occurs at the stage in 
which voiceP is built. The proposal that I will now adopt for middles is taken from Bruening 
(2012, 29–31).
Bruening suggests that the derivation of a middle sentence proceeds initially in the same manner 
as that of a prototypical active transitive structure, in the sense that a vP containing the internal 
argument first merges with a voice head introducing the open agent variable. The point of 
syntactic departure in the middle sentence, however, happens because it merges a semantic 
operator that selects an unsaturated voice projection (i.e., the denotation of the intermediate 
category Voice’ in (19)), and thereby blocks the merger of a referring DP as the external argument. 
The semantics for such an operator are given in (21).
(21) 
To derive a prototypical transitive sentence such as John read the book, what the syntactic 
computation needs to do is simply merge the DP John in the [Spec,VoiceP] position to saturate 
the open variable of Voice’ (20b). The final semantics for VoiceP is given in (20c).
(20) The semantics of VoiceP with external argument John
a. ⟦VoiceP ⟧= ⟦Voice’ ⟧( ⟦John ⟧) = 
b. λx ⟨e ⟩ .λe ⟨s ⟩.[AGENT(x, e) ∧ READ(e) ∧ PATIENT(the book, e)](John) = 
c. λe ⟨s ⟩.[AGENT(John, e) ∧ READ(e) ∧ PATIENT(the book, e)]
2.3 A Syntactic Account of English Middles
There is an agreement in the syntactic literature (e.g., Bruening 2012, Legate 2014, Alexiadou,
Anagnostopoulou, and Schäfer 2015) that VP-external valency reduction (that is, the omission
of an overt external argument in middles, anticausatives, passives, etc.) occurs at the stage in 
which VoiceP is built. The proposal that I will now adopt for middles is taken from Bruening
(2012, 29–31).
Bruening suggests that the derivation of a middle sentence proceeds initially in the same
manner as that of a prototypical active transitive structure, in the sense that a VP containing 
the internal argument first merges with a Voice head introducing the open agent variable. The 
point of syntactic departure in the middle sentence, however, happens because it merges a
semantic operator that selects an unsaturated Voice projection (i.e., the denotation of the
intermediate category Voice’ in (19)), and thereby blocks the merger of a referring DP as the 
external argument. The semantics for such an operator are given in (21).
(21) ⟦MiddleOp ⟧ = λf ⟨e, ⟨s,t⟩⟩ .λ g ⟨s,t ⟩.GNe ∃x[f(x, e) → g(e)]
(Bruening 2012, 31)
According to (21), MiddleOp selects as its first semantic argument a syntactic category that 
logically corresponds to a function with an open agent variable, and maps it onto a scope where
the agent variable is bound by an existential quantifier. The second argument fed into (21) is 
an event modifier of type ⟨s,t ⟩, which corresponds to the adverbial modification in middles 
(i.e., an adverb like easily is interpreted as a function from easy events to truth values; see the 
syntactic tree in (22) below).
Syntactically, note that Bruening’s (2012) proposal is cast within a theory of bare-phrase 
structure where operators can be merged before a grammatical head – Voice, in this case – fully
projects. Differently put, the semantics in (21) suggests that MiddleOp is merged in the [Spec,
VoiceP] position, but this position is canonically (e.g., in X-bar theory, whose notation the
paper has partially adopted) associated with nominal expressions (DPs saturating the agent
variable), and not with operators that select their own semantic arguments. Although such an
assumption is empirically and theoretically well motivated (see the discussion in Bruening 
                                                                                          (Bruening 2012, 31)
According to (21), MiddleOp selects as its first semantic argument a syntactic category that 
logically corresponds to a function with an open agent variable, and maps it onto a scope where 
the agent variable is bound by an existential quantifier. The second argument fed into (21) is an 
event modifier of type ⟨s,t⟩, which corresponds to the adverbial modification in middles (i.e., an 
adverb like easily is interpreted as a function from easy events to truth values; see the syntactic 
tree in (22) below).
where they combine via Functional Application with the predicate in the same manner as the simple definite descriptor, 
which is of type ⟨ e ⟩, in (18c).

51 LANGUAGE
Syntactically, note that Bruening’s (2012) proposal is cast within a theory of bare-phrase structure 
where operators can be merged before a grammatical head – v oice, in this case – fully projects. 
Differently put, the semantics in (21) suggests that MiddleOp is merged in the [Spec, voiceP] 
position, but this position is canonically (e.g., in X-bar theory, whose notation the paper has 
partially adopted) associated with nominal expressions (DPs saturating the agent variable), and 
not with operators that select their own semantic arguments. Although such an assumption is 
empirically and theoretically well motivated (see the discussion in Bruening 2012, 27–28), it 
seems that the whole derivation can be simplified – at least for ease of exposition – by assuming 
that MiddleOp heads its own phrase and selects a voiceP which has not merged an external 
argument in its specifier slot. That is, the vP of a middle construction merges with a special 
v oice head, which introduces an open agent variable, but does not require a filled specifier slot.
3
 
I thus propose that an English middle construction has the syntactic representation in (22).
(22) 
2012, 27–28), it seems that the whole derivation can be simplified – at least for ease of 
exposition – by assuming that MiddleOp heads its own phrase and selects a VoiceP which has 
not merged an external argument in its specifier slot. That is, the VP of a middle construction 
merges with a special Voice head, which introduces an open agent variable, but does not require 
a filled specifier slot.
3
I thus propose that an English middle construction has the syntactic 
representation in (22).
(22)
Applying MiddleOp to the specifier-less Voice projection, the syntactic derivation arrives at
the semantic representation of the middle sentence This book reads easily that is given in (23).
(23) The interpretation of This book reads easily
a. ⟦MiddleOp ⟧( ⟦VoiceP ⟧)( ⟦easily⟧) =
b. λf ⟨e, ⟨s,t⟩⟩.λg ⟨s,t ⟩.GNe ∃x[f(x, e) → g(e)](λx ⟨e ⟩.λe ⟨s ⟩.[AGENT(x, e) ∧ READ(e) ∧
PATIENT(the book, e)]) (λe ⟨s ⟩.[EASY(e)]) =
c. GNe ∃x[[AGENT(x, e) ∧ READ(e) ∧ PATIENT(the book, e)] → [EASY(e)]]
Since the existentially bound agent variable is within the scope of a generic quantifier, i.e. GN, 
the final interpretation of a sentence such as This book reads easily, given in (23c), is 
Generally, whenever someone reads the book, she does so easily.
Such a semantics and its corresponding syntactic derivation account for many of the properties 
of middles discussed in Section 2.1. First, recall that, cross-linguistically, middles do not 
pattern morphosyntactically with unaccusative sentences, which merge their sole argument 
VP-internally, but rather with unergative sentences, which merge their argument VP-
externally, in that they select the auxiliary have instead of be in periphrastic tense constructions 
in languages such as Dutch (examples (14)–(15)). On the current account, the patterning with 
unergatives simply follows from the fact that middles contain a Voice projection whose head
semantically brings about the external participant. Such a VoiceP is also present in unergatives 
and transitives, but not in unaccusatives, which is why they select be. Furthermore, note that 
the English middle construction can only be headed by a lexically transitive predicate (24a), in 
contrast with a prototypical unaccusative sentence (24b).
(24) a. Obedient daughters raise/*rise easily.
                                                
3
This assumption is in line with recent work on the grammatical properties of the Voice domain (e.g., Legate 
2014; Alexiadou, Anagnostopoulou, and Schäfer 2015), whose central claim is that languages employ a set of 
different Voice heads with unique selectional (e.g., a Voice head may or may not introduce the external participant 
variable) and syntactic (e.g., a Voice projection may or may not require a filled specifier slot) properties.
Applying MiddleOp to the specifier-less v oice projection, the syntactic derivation arrives at the 
semantic representation of the middle sentence This book reads easily that is given in (23).
(23) The interpretation of This book reads easily
 
2012, 27–28), it seems that the whole derivation can be simplified – at least for ease of 
exposition – by assuming that MiddleOp heads its own phrase and selects a VoiceP which has 
not merged an external argument in its specifier slot. That is, the VP of a middle construction 
merges with a special Voice head, which introduces an open agent variable, but does not require 
a filled specifier slot.
3
I thus propose that an English middle construction has the syntactic 
representation in (22).
(22)
Applying MiddleOp to the specifier-less Voice projection, the syntactic derivation arrives at
the semantic representation of the middle sentence This book reads easily that is given in (23).
(23) The interpretation of This book reads easily
a. ⟦MiddleOp ⟧( ⟦VoiceP ⟧)( ⟦easily⟧) =
b. λf ⟨e, ⟨s,t⟩⟩.λg ⟨s,t ⟩.GNe ∃x[f(x, e) → g(e)](λx ⟨e⟩.λe ⟨s ⟩.[AGENT(x, e) ∧ READ(e) ∧
PATIENT(the book, e)]) (λe ⟨s ⟩.[EASY(e)]) =
c. GNe ∃x[[AGENT(x, e) ∧ READ(e) ∧ PATIENT(the book, e)] → [EASY(e)]]
Since the existentially bound agent variable is within the scope of a generic quantifier, i.e. GN, 
the final interpretation of a sentence such as This book reads easily, given in (23c), is 
Generally, whenever someone reads the book, she does so easily.
Such a semantics and its corresponding syntactic derivation account for many of the properties 
of middles discussed in Section 2.1. First, recall that, cross-linguistically, middles do not 
pattern morphosyntactically with unaccusative sentences, which merge their sole argument 
VP-internally, but rather with unergative sentences, which merge their argument VP-
externally, in that they select the auxiliary have instead of be in periphrastic tense constructions 
in languages such as Dutch (examples (14)–(15)). On the current account, the patterning with 
unergatives simply follows from the fact that middles contain a Voice projection whose head
semantically brings about the external participant. Such a VoiceP is also present in unergatives 
and transitives, but not in unaccusatives, which is why they select be. Furthermore, note that 
the English middle construction can only be headed by a lexically transitive predicate (24a), in 
contrast with a prototypical unaccusative sentence (24b).
(24) a. Obedient daughters raise/*rise easily.
                                                
3
This assumption is in line with recent work on the grammatical properties of the Voice domain (e.g., Legate 
2014; Alexiadou, Anagnostopoulou, and Schäfer 2015), whose central claim is that languages employ a set of 
different Voice heads with unique selectional (e.g., a Voice head may or may not introduce the external participant 
variable) and syntactic (e.g., a Voice projection may or may not require a filled specifier slot) properties.
Since the existentially bound agent variable is within the scope of a generic quantifier, i.e. GN, 
the final interpretation of a sentence such as This book reads easily, given in (23c), is Generally, 
whenever someone reads the book, she does so easily. 
Such a semantics and its corresponding syntactic derivation account for many of the properties 
of middles discussed in Section 2.1. First, recall that, cross-linguistically, middles do not 
pattern morphosyntactically with unaccusative sentences, which merge their sole argument vP-
internally, but rather with unergative sentences, which merge their argument vP-externally, in 
that they select the auxiliary have instead of be in periphrastic tense constructions in languages 
such as Dutch (examples (14)–(15)). On the current account, the patterning with unergatives 
simply follows from the fact that middles contain a voice projection whose head semantically 
brings about the external participant. Such a v oiceP is also present in unergatives and transitives, 
3
 This assumption is in line with recent work on the grammatical properties of the v oice domain (e.g., Legate 2014; Alexiadou, 
Anagnostopoulou, and Schäfer 2015), whose central claim is that languages employ a set of different voice heads with 
unique selectional (e.g., a v oice head may or may not introduce the external participant variable) and syntactic (e.g., a v oice 
projection may or may not require a filled specifier slot) properties.

52 Jakob Lenardič  The Middle Construction in Minimalist Syntax: English and Slovenian
but not in unaccusatives, which is why they select be. Furthermore, note that the English middle 
construction can only be headed by a lexically transitive predicate (24a), in contrast with a 
prototypical unaccusative sentence (24b).
(24) a. Obedient daughters raise/*rise easily.
   b. The sun rises/*raises in the east. 
                                                           (Examples from Schäfer 2007)
An additional advantage of this syntactic approach is that it does not need to posit a special 
lexical operation whereby the external participant in the middle is realised as an arbitrary agent 
(Fagan 1992; Ackema and Schoorlemmer 1994; Lekakou 2004; Marelj 2004). The arbitrary 
flavour simply comes from the fact that the existentially bound agent variable x occurs within 
the scope of the generic operator (which binds a situation variable); that is, the final LF of a  
middle sentence in (23c), which corresponds to the paraphrase Generally, whenever someone reads 
the book, she does so easily, is interpretatively very similar to Fagan’s (1992) proposed paraphrase 
Anyone can read this book easily, in the sense that “anyone” is simply “someone” who differs from 
one generic situation to another. conceptually, then, such a syntactic account allows us to derive 
a middle sentence without recourse to a transformative phase in the lexicon that pre-syntactically 
turns a verb like read, which is inherently transitive, into a middle variant with an independently 
posited arbitrary agent role (cf. (12)–(13)).
The greatest advantage of the derivational syntactic account, however, is that it offers a principled 
formal explanation as to why a middle sentence licenses an instrumental with-phrase. Recall 
that, in this respect, the middle sentence in (25a) patterns with the regular active sentence in 
(25b), but not with the anticausative structure in (25c) formed from the same predicate.
(25) The licensing of an instrumental with-phrase
  a. Middle: This door opens easily with a key.
  b. Active: John opened this door with a key.
  c. Anticausative: The door opened (*with a key).
Bruening (2012, 27) proposes that with is interpretatively a function that selects the denotation 
of its syntactic complement DP , which denotes some instrument used by the agent, as its first 
semantic argument.
(26) 
b. The sun rises/*raises in the east.
(Examples from Schäfer 2007)
An additional advantage of this syntactic approach is that it does not need to posit a special
lexical operation whereby the external participant in the middle is realised as an arbitrary agent
(Fagan 1992; Ackema and Schoorlemmer 1994; Lekakou 2004; Marelj 2004). The arbitrary 
flavour simply comes from the fact that the existentially bound agent variable x occurs within 
the scope of the generic operator (which binds a situation variable); that is, the final LF of a  
middle sentence in (23c), which corresponds to the paraphrase Generally, whenever someone 
reads the book, she does so easily, is interpretatively very similar to Fagan’s (1992) proposed 
paraphrase Anyone can read this book easily, in the sense that “anyone” is simply “someone” 
who differs from one generic situation to another. Conceptually, then, such a syntactic account 
allows us to derive a middle sentence without recourse to a transformative phase in the lexicon
that pre-syntactically turns a verb like read, which is inherently transitive, into a middle variant 
with an independently posited arbitrary agent role (cf. (12)–(13)).
The greatest advantage of the derivational syntactic account, however, is that it offers a 
principled formal explanation as to why a middle sentence licenses an instrumental with-
phrase. Recall that, in this respect, the middle sentence in (25a) patterns with the regular active
sentence in (25b), but not with the anticausative structure in (25c) formed from the same 
predicate.
(25) The licensing of an instrumental with-phrase
a. Middle: This door opens easily with a key.
b. Active: John opened this door with a key.
c. Anticausative: The door opened (*with a key).
Bruening (2012, 27) proposes that with is interpretatively a function that selects the denotation 
of its syntactic complement DP, which denotes some instrument used by the agent, as its first 
semantic argument.
(26) ⟦with ⟧ = λx ⟨e ⟩ .λf ⟨e, ⟨s,t⟩⟩.λy ⟨e ⟩.λe ⟨s ⟩.[f(y, e) ∧ ∃ e’≤ e[USING(x, e’) ∧ AGENT(y, e’)]]
Let us assume that there is an event e in which someone opens the door. The semantics in (26)
say that there exists a subevent of e – namely e’ –, in which a person interpreted as an agent y
opens the door by using an instrument x. In other words, what the preposition with does
semantically is it expands the denotation of the syntactic structure by introducing an additional
participant that interpretatively corresponds to the instrument.
Syntactically, the key idea behind (26) is that a with-phrase can only attach to a structure which 
contains an open agent variable, as seen in the fact that with obligatorily selects as its second
semantic argument a function of type ⟨e, ⟨s,t ⟩⟩. On the syntactic account, middles are indeed 
built on top of such a structure (VoiceP in (22)), hence the reason why a with-phrase is 
admissible. By contrast, an anticausative structure such as (25c) is standardly analysed as either 
omitting a Voice projection or containing one that is semantically vacuous, which means that 
the denotation of VP is not expanded to contain an agent variable.
4
The denotation of a non-
                                                
4
According to Schäfer and Vivanco (2016), an anticausative sentence like This door opened simply means There 
was an event in which this door became open; since Voice is either not merged or semantically inert, no specific 
type of external participation is entailed. The syntactic and semantic lack of an external participant makes the 
sentence compatible with describing a number of possible situations in which the door becomes open; the external 
participant might be a non-human causer (e.g., The wind opened the door), or there even might not be any apparent 
Let us assume that there is an event e in which someone opens the door. The semantics in (26) 
say that there exists a subevent of e – namely e’ –, in which a person interpreted as an agent 
y opens the door by using an instrument x. In other words, what the preposition with does 
semantically is it expands the denotation of the syntactic structure by introducing an additional 
participant that interpretatively corresponds to the instrument. 
Syntactically, the key idea behind (26) is that a with-phrase can only attach to a structure which 
contains an open agent variable, as seen in the fact that with obligatorily selects as its second 
semantic argument a function of type ⟨e,⟨s,t⟩⟩ . On the syntactic account, middles are indeed built 
on top of such a structure ( v oiceP in (22)), hence the reason why a with-phrase is admissible. By 
contrast, an anticausative structure such as (25c) is standardly analysed as either omitting a v oice 

53 LANGUAGE
projection or containing one that is semantically vacuous, which means that the denotation 
of vP is not expanded to contain an agent variable.
4
 The denotation of a non-expanded vP is 
of type ⟨s,t⟩ (a function from eventualities to truth values), as in (18c), so an instrumental PP 
cannot be structurally adjoined to it, as it would result in a type mismatch.
I believe that this is, from the point of view of minimalist theory, the best way in which the 
Instrument Generalization, which states that “an [instrumental PP] requires the explicit (syntactic) 
or implicit (semantic) presence of an agent in order to be realized syntactically” (Marelj 2004, 
116), can be accounted for in a formal manner.  The key advantage of the syntactic approach 
is that it re-states this rather informal – and quite vague – distinction between an agent that 
is present only implicitly and one that is present explicitly in much clearer terms: the implicit 
agent in a middle is one that begins its grammatical life as an open variable denoting a volitional 
participant, only to be later on saturated by existential closure brought about by an operator such 
as (21). Hence, because the agent is at one point an open variable, the with-phrase is admitted.
3 The Role of the Adverbial Modifier 
In this section, I propose an amendment to Bruening’s (2012) account related to the adverbial 
modifier, which is posited as a necessary requirement of the middle construction because 
MiddleOp introduces the generic operator GN. According to the literature on genericity (Krifka 
et al. 1995), a generic quantifier GN partitions the LF of a generic sentence into two parts – 
the restrictive and the nuclear scope. It is a necessary condition of such a two-part LF that the 
nuclear scope have semantic content (Krifka et al. 1995, 25–27), which explains why a sentence 
like This book reads *(easily) is ill-formed if easily is omitted. 
c rucially, the semantics of MiddleOp currently predict that the LF partitioning is taken care of 
by the syntactic derivation itself; namely, by the fact that the operator requires an AdvP as its 
second semantic argument, which it then maps into the nuclear scope. I will now show that such 
a requirement is too strict to account for all English middles, and that the formal representation of 
MiddleOp should be simplified from its original formulation in (21), repeated here in (27), to (28).
(27) 
expanded VP is of type ⟨s,t ⟩ (a function from eventualities to truth values), as in (18c), so an
instrumental PP cannot be structurally adjoined to it, as it would result in a type mismatch.
I believe that this is, from the point of view of minimalist theory, the best way in which the
Instrument Generalization, which states that “an [instrumental PP] requires the explicit 
(syntactic) or implicit (semantic) presence of an agent in order to be realized syntactically” 
(Marelj 2004, 116), can be accounted for in a formal manner.  The key advantage of the 
syntactic approach is that it re-states this rather informal – and quite vague – distinction 
between an agent that is present only implicitly and one that is present explicitly in much clearer
terms: the implicit agent in a middle is one that begins its grammatical life as an open variable
denoting a volitional participant, only to be later on saturated by existential closure brought 
about by an operator such as (21). Hence, because the agent is at one point an open variable, 
the with-phrase is admitted.
3 The Role of the Adverbial Modifier 
In this section, I propose an amendment to Bruening’s (2012) account related to the adverbial
modifier, which is posited as a necessary requirement of the middle construction because 
MiddleOp introduces the generic operator GN. According to the literature on genericity (Krifka 
et al. 1995), a generic quantifier GN partitions the LF of a generic sentence into two parts – the 
restrictive and the nuclear scope. It is a necessary condition of such a two-part LF that the 
nuclear scope have semantic content (Krifka et al. 1995, 25–27), which explains why a sentence
like This book reads *(easily) is ill-formed if easily is omitted.
Crucially, the semantics of MiddleOp currently predict that the LF partitioning is taken care of 
by the syntactic derivation itself; namely, by the fact that the operator requires an AdvP as its 
second semantic argument, which it then maps into the nuclear scope. I will now show that 
such a requirement is too strict to account for all English middles, and that the formal 
representation of MiddleOp should be simplified from its original formulation in (21), repeated 
here in (27), to (28).
(27) ⟦MiddleOp ⟧ (original version) = λf ⟨e, ⟨s,t⟩⟩.λg ⟨s,t ⟩.GNe ∃x[f(x, e) → g(e)]
(28) ⟦MiddleOp ⟧ (revised version)  = λf ⟨e, ⟨s,t⟩⟩.GNe ∃x[f(x, e)]
What is key to the simplified semantics in (28) is that the adverbial modifier, although 
obligatory in many middles, is no longer a semantic argument of the operator itself. There are 
two pieces of evidence for this assumption and the consequent revision. First, there are middle 
sentences in English that are perfectly licit without an AdvP, as shown by the following two 
examples, taken from Marelj (2004, 125).
(29) This umbrella folds up.
(30) This dress buttons.
Despite the absence of the AdvP, the LF of such sentences is still partitioned into a restrictive 
and nuclear scope, as required by the generic quantifier. According to Marelj (2004, 127–128),
the LF partitioning has an extra-semantic source in contrast to the original semantics of 
MiddleOp in (27) – i.e., it is licensed by prosodic structure, in that “the focused part of an 
utterance with an operator such as [GN] is always in the [nuclear scope]” (Krifka et al. 1995, 
27). What then licenses an adverb-less middle sentence like example (30) is the placement of 
focus stress on the verb buttons (McConnell-Ginet 1994). Such focus placement brings about
                                                
external cause, as is evident by the fact that an anticausative sentence admits the by-itself phrase: The door opened 
by itself.
 
(28) 
expanded VP is of type ⟨s,t ⟩ (a function from eventualities to truth values), as in (18c), so an
instrumental PP cannot be structurally adjoined to it, as it would result in a type mismatch.
I believe that this is, from the point of view of minimalist theory, the best way in which the
Instrument Generalization, which states that “an [instrumental PP] requires the explicit 
(syntactic) or implicit (semantic) presence of an agent in order to be realized syntactically” 
(Marelj 2004, 116), can be accounted for in a formal manner.  The key advantage of the 
syntactic approach is that it re-states this rather informal – and quite vague – distinction 
between an agent that is present only implicitly and one that is present explicitly in much clearer
terms: the implicit agent in a middle is one that begins its grammatical life as an open variable
denoting a volitional participant, only to be later on saturated by existential closure brought 
about by an operator such as (21). Hence, because the agent is at one point an open variable, 
the with-phrase is admitted.
3 The Role of the Adverbial Modifier 
In this section, I propose an amendment to Bruening’s (2012) account related to the adverbial
modifier, which is posited as a necessary requirement of the middle construction because 
MiddleOp introduces the generic operator GN. According to the literature on genericity (Krifka 
et al. 1995), a generic quantifier GN partitions the LF of a generic sentence into two parts – the 
restrictive and the nuclear scope. It is a necessary condition of such a two-part LF that the 
nuclear scope have semantic content (Krifka et al. 1995, 25–27), which explains why a sentence
like This book reads *(easily) is ill-formed if easily is omitted.
Crucially, the semantics of MiddleOp currently predict that the LF partitioning is taken care of 
by the syntactic derivation itself; namely, by the fact that the operator requires an AdvP as its 
second semantic argument, which it then maps into the nuclear scope. I will now show that 
such a requirement is too strict to account for all English middles, and that the formal 
representation of MiddleOp should be simplified from its original formulation in (21), repeated 
here in (27), to (28).
(27) ⟦MiddleOp ⟧ (original version) = λf ⟨e, ⟨s,t⟩⟩.λg ⟨s,t ⟩.GNe ∃x[f(x, e) → g(e)]
(28) ⟦MiddleOp ⟧ (revised version)  = λf ⟨e, ⟨s,t⟩⟩.GNe ∃x[f(x, e)]
What is key to the simplified semantics in (28) is that the adverbial modifier, although 
obligatory in many middles, is no longer a semantic argument of the operator itself. There are 
two pieces of evidence for this assumption and the consequent revision. First, there are middle 
sentences in English that are perfectly licit without an AdvP, as shown by the following two 
examples, taken from Marelj (2004, 125).
(29) This umbrella folds up.
(30) This dress buttons.
Despite the absence of the AdvP, the LF of such sentences is still partitioned into a restrictive 
and nuclear scope, as required by the generic quantifier. According to Marelj (2004, 127–128),
the LF partitioning has an extra-semantic source in contrast to the original semantics of 
MiddleOp in (27) – i.e., it is licensed by prosodic structure, in that “the focused part of an 
utterance with an operator such as [GN] is always in the [nuclear scope]” (Krifka et al. 1995, 
27). What then licenses an adverb-less middle sentence like example (30) is the placement of 
focus stress on the verb buttons (McConnell-Ginet 1994). Such focus placement brings about
                                                
external cause, as is evident by the fact that an anticausative sentence admits the by-itself phrase: The door opened 
by itself.
What is key to the simplified semantics in (28) is that the adverbial modifier, although obligatory 
in many middles, is no longer a semantic argument of the operator itself. There are two pieces 
of evidence for this assumption and the consequent revision. First, there are middle sentences in 
English that are perfectly licit without an AdvP , as shown by the following two examples, taken 
from Marelj (2004, 125).
(29) This umbrella folds up.
(30) This dress buttons.
4
 According to Schäfer and v ivanco (2016), an anticausative sentence like This door opened simply means There was an event in 
which this door became open; since Voice is either not merged or semantically inert, no specific type of external participation is 
entailed. The syntactic and semantic lack of an external participant makes the sentence compatible with describing a number 
of possible situations in which the door becomes open; the external participant might be a non-human causer (e.g., The wind 
opened the door), or there even might not be any apparent external cause, as is evident by the fact that an anticausative sentence 
admits the by-itself phrase: The door opened by itself.

54 Jakob Lenardič  The Middle Construction in Minimalist Syntax: English and Slovenian
Despite the absence of the AdvP , the LF of such sentences is still partitioned into a restrictive and 
nuclear scope, as required by the generic quantifier. According to Marelj (2004, 127–128), the 
LF partitioning has an extra-semantic source in contrast to the original semantics of MiddleOp 
in (27) – i.e., it is licensed by prosodic structure, in that “the focused part of an utterance with 
an operator such as [GN] is always in the [nuclear scope]” (Krifka et al. 1995, 27). What then 
licenses an adverb-less middle sentence like example (30) is the placement of focus stress on the 
verb buttons (Mcconnell-Ginet 1994). Such focus placement brings about the LF in (31), where 
(a part of) the denotation of the verb buttons is mapped into the nuclear scope. 
(31) 
the LF in (31), where (a part of) the denotation of the verb buttons is mapped into the nuclear 
scope. 
 GNe∃x[[ ∧ gent (x, e) ∧ → The LF in (31), in which C is taken to be a variable ranging over contextually relevant 
information, can be paraphrased as Whenever someone does something to this dress in a 
contextually determined event, she buttons it, where the contextually determined event is “most 
likely associated with some aspect of x putting on a dress in an appropriate way” (Marelj 2004, 
128).
5
Such an LF, however, is not predicted by the original formulation of MiddleOp in (27), 
which banks on the assumption that the LF partition necessary requires the syntactic presence 
of an additional constituent (i.e., the adverbial modifier), which is then mapped into the nuclear 
scope.  In other words, this assumption goes against the adverb-less middle example (30), as 
the partitioning of its LF takes place without the additional piece of structure. 
The second piece of evidence lies in the fact that the operator’s requirement for a second 
semantic argument derives word order that is too rigid. To see this, recall from Section 2.3 that 
an instrumental with-phrase enters the derivation as a Voice adjunct. What is crucial is that this 
takes place before MiddleOp enters the structure, given that with semantically needs an open 
agent variable (26). Consequently, if MiddleOp had the semantics in (27), the word order would 
be invariably as in (32); namely, the AdvP would necessarily enter the structure higher than 
the with-phrase to satisfy the second semantic requirement of the operator, making the standard 
word order in (33) difficult, if not impossible, to derive.
(32) ?This door opens with a key easily. 
(33) This door opens easily with a key.
Additionally, it has been observed by Fellbaum (1986) that the AdvP cannot precede the verb 
in the middle construction (34), while such word order is possible in a non-middle anticausative 
structure (35):
(34) This book (*easily) reads (easily).
(35) Yesterday, the door (quickly) opened (quickly).
Schäfer (2007) claims that the difference in the potential placement of the AdvP follows from 
the fact that (34) is structurally richer than (35) (that is, it contains VoiceP) and from the 
additional assumption that the verb moves to the Voice head, as in (36a). By contrast, the 
anticausative sentence in (35) lacks the Voice layer, so the verb remains in situ (36b), and the 
adverb is able to precede it.
(36) a. [TP this book [T’ T [VoiceP [Voice’ Voice+reads [VP easily [VP reads this book]]]]]]
b. [TP this door [T’ T [VP quickly [VP opened the door]]]]
This fact can only be accounted for if the adverb is merged low in the structure as a VP adjunct, 
so before MiddleOp even enters the syntactic derivation. What follows, then, is that the adverb 
is not a semantic requirement of the operator itself. It is, however, needed in a sentence like 
This book reads easily by the generic quantifier because of the well-formedness condition on 
its bipartite scope structure (i.e., the nuclear scope mustn’t be empty and must contain non-
trivial information, cf. footnote 5). Crucially, this requirement on part of the quantifier is not 
                                                
5
In this sense, a sentence like *This book READS is still bad without the adverb because placing focus on reads
leads to trivial information getting mapped into the nuclear scope, as all books are such that one can read them, 
whereas not all dresses are such that one can button them (Marelj 2004).  
 (Adapted from Marelj 2004, 127) 
The LF in (31), in which c is taken to be a variable ranging over contextually relevant information, 
can be paraphrased as Whenever someone does something to this dress in a contextually determined 
event, she buttons it, where the contextually determined event is “most likely associated with 
some aspect of x putting on a dress in an appropriate way” (Marelj 2004, 128).
5
 Such an LF, 
however, is not predicted by the original formulation of MiddleOp in (27), which banks on 
the assumption that the LF partition necessarily requires the syntactic presence of an additional 
constituent (i.e., the adverbial modifier), which is then mapped into the nuclear scope.  In other 
words, this assumption goes against the adverb-less middle example (30), as the partitioning of 
its LF takes place without the additional piece of structure. 
The second piece of evidence lies in the fact that the operator’s requirement for a second 
semantic argument derives word order that is too rigid. to see this, recall from Section 2.3 that 
an instrumental with-phrase enters the derivation as a v oice adjunct. What is crucial is that this 
takes place before MiddleOp enters the structure, given that with semantically needs an open 
agent variable (26). consequently, if MiddleOp had the semantics in (27), the word order would 
be invariably as in (32); namely, the AdvP would necessarily be higher in the structure than the 
with-phrase because of the second semantic requirement of the operator, making the standard 
word order in (33) difficult, if not impossible, to derive.
(32)  ?This door opens with a key easily. 
(33)  This door opens easily with a key.
Additionally, it has been observed by Fellbaum (1986) that the AdvP cannot precede the verb in 
the middle construction (34), while such word order is possible in a non-middle anticausative 
structure (35):
(34) This book (*easily) reads (easily).
(35) Yesterday, the door (quickly) opened (quickly).
Schäfer (2007) claims that the difference in the potential placement of the AdvP follows from the 
fact that (34) is structurally richer than (35) (that is, it contains voiceP) and from the additional 
assumption that the verb moves to the v oice head, as in (36a). By contrast, the anticausative sentence 
in (35) lacks the v oice layer, so the verb remains in situ (36b), and the adverb is able to precede it.
5
 In this sense, a sentence like *This book READS is still bad without the adverb because placing focus on reads leads to trivial 
information getting mapped into the nuclear scope, as all books are such that one can read them, whereas not all dresses are 
such that one can button them (Marelj 2004).  

55 LANGUAGE
(36) 
the LF in (31), where (a part of) the denotation of the verb buttons is mapped into the nuclear 
scope. 
 GNe∃x[[ ∧ gent (x, e) ∧ → The LF in (31), in which C is taken to be a variable ranging over contextually relevant 
information, can be paraphrased as Whenever someone does something to this dress in a 
contextually determined event, she buttons it, where the contextually determined event is “most 
likely associated with some aspect of x putting on a dress in an appropriate way” (Marelj 2004, 
128).
5
Such an LF, however, is not predicted by the original formulation of MiddleOp in (27), 
which banks on the assumption that the LF partition necessary requires the syntactic presence 
of an additional constituent (i.e., the adverbial modifier), which is then mapped into the nuclear 
scope.  In other words, this assumption goes against the adverb-less middle example (30), as 
the partitioning of its LF takes place without the additional piece of structure. 
The second piece of evidence lies in the fact that the operator’s requirement for a second 
semantic argument derives word order that is too rigid. To see this, recall from Section 2.3 that 
an instrumental with-phrase enters the derivation as a Voice adjunct. What is crucial is that this 
takes place before MiddleOp enters the structure, given that with semantically needs an open 
agent variable (26). Consequently, if MiddleOp had the semantics in (27), the word order would 
be invariably as in (32); namely, the AdvP would necessarily enter the structure higher than 
the with-phrase to satisfy the second semantic requirement of the operator, making the standard 
word order in (33) difficult, if not impossible, to derive.
(32) ?This door opens with a key easily. 
(33) This door opens easily with a key.
Additionally, it has been observed by Fellbaum (1986) that the AdvP cannot precede the verb 
in the middle construction (34), while such word order is possible in a non-middle anticausative 
structure (35):
(34) This book (*easily) reads (easily).
(35) Yesterday, the door (quickly) opened (quickly).
Schäfer (2007) claims that the difference in the potential placement of the AdvP follows from 
the fact that (34) is structurally richer than (35) (that is, it contains VoiceP) and from the 
additional assumption that the verb moves to the Voice head, as in (36a). By contrast, the 
anticausative sentence in (35) lacks the Voice layer, so the verb remains in situ (36b), and the 
adverb is able to precede it.
(36) a. [TP this book [T’ T [VoiceP [Voice’ Voice+reads [VP easily [VP reads this book]]]]]]
b. [TP this door [T’ T [VP quickly [VP opened the door]]]]
This fact can only be accounted for if the adverb is merged low in the structure as a VP adjunct, 
so before MiddleOp even enters the syntactic derivation. What follows, then, is that the adverb 
is not a semantic requirement of the operator itself. It is, however, needed in a sentence like 
This book reads easily by the generic quantifier because of the well-formedness condition on 
its bipartite scope structure (i.e., the nuclear scope mustn’t be empty and must contain non-
trivial information, cf. footnote 5). Crucially, this requirement on part of the quantifier is not 
                                                
5
In this sense, a sentence like *This book READS is still bad without the adverb because placing focus on reads
leads to trivial information getting mapped into the nuclear scope, as all books are such that one can read them, 
whereas not all dresses are such that one can button them (Marelj 2004).  
 
 (The structures in (36) are modelled after Schäfer 2007.)
This fact can only be accounted for if the adverb is merged low in the structure as a vP adjunct, 
so before MiddleOp even enters the syntactic derivation. What follows, then, is that the adverb 
is not a semantic requirement of the operator itself. It is, however, needed in a sentence like 
This book reads easily by the generic quantifier because of the well-formedness condition on its 
bipartite scope structure (i.e., the nuclear scope mustn’t be empty and must contain non-trivial 
information, cf. footnote 5). c rucially, this requirement on part of the quantifier is not built 
into the syntactic component, as predicted by the revised semantics of MiddleOp in (28), but 
holds at LF, given that it can be satisfied in middles that lack adverbs (examples (29)-(30)) by 
non-semantic means (i.e., focus placement).
4 The Slovenian se-Construction
In this section, I turn to some of the main grammatical characteristics of the Slovenian middle 
construction and contrast them with those of the corresponding English construction. Recall from 
Section 1 that Slovenian middles differ from English middles in two key ways. First, they contain 
an additional syntactic category in the form of the reflexive morpheme se (37).
6
 Second, they allow 
a non-generic interpretation (38), which is not admissible in the English construction (39).
(37)     Ta   knjiga       se      bere         zlahka.
       this book.nom refl  read.3sg     easily 
      “This book reads easily.”
(38)     Ta knjiga                se      je           včeraj        brala.
        this book.nom.fem refl aux.3sg yesterday       read.l-ptcpl. fem
      “Yesterday, this book was read.”
(39)    *Yesterday, this book read.
     Intended: “Yesterday, someone read this book.”
The general intuition in the formal literature is that se is the source of cross-linguistic differences; 
as Alexiadou (2010, 3) succinctly proposes, “the morphology we see in the alternation should be 
taken seriously and is the device that helps us explain why [superficial valency reduction] is freer 
in some languages than others”. Hence, it is the lack of se in English middles that limits their 
distribution in comparison to those in Slovenian. 
6
 Note that sentences (37) and (38) are similar to the Slovenian impersonal se-construction, in which the external argument 
DP (denoting the reader of the book) is also superficially absent from the overt structural representation, as in the following 
example, which is interpretatively equivalent to (37).
(i) to knjigo       se     zlahka bere.
 this book.acc refl easily  read.
 “This book reads easily.”
Rivero and Milojević Sheppard (2003) argue at length that an impersonal sentence such as (i) is a two-argument structure; in 
addition to the overt internal argument DP (T o knjigo), the sentence contains a thematic agent argument, which is realised as 
a phonologically null pronoun. By contrast, sentences (37) and (38) contain only one syntactic thematic argument (i.e., the 
internal one), whereas the agent is present only semantically as an existentially bound agent variable; cf. section 4.2 below. 

56 Jakob Lenardič  The Middle Construction in Minimalist Syntax: English and Slovenian
4.1 The Syntax of se 
As regards the syntactic role of se, I will adopt a recent proposal by Alexiadou, Anagnostopoulou, 
and Schäfer (2015, chapter 4) and Schäfer (2017), in which a reflexive morpheme like se is 
analysed as a nominal DP that occupies the specifier position of a v oiceP , as in (40).
7
 
(40) The VoiceP of a se-sentence (Schäfer 2017)
The core syntactic idea underlying (40) is that a sentence with se is built up in the same manner 
as a prototypical active transitive construction in terms of argument structure, in the sense that 
it contains both an internal argument in the VP, knjiga, as well as an external argument in the 
specifier of VoiceP; that is, se. At first look, the only thing that distinguishes a se-sentence like 
(38), Ta knjigaNOM se je včeraj brala, from a prototypical active transitive sentence such as 
MarijaNOM je včeraj brala to knjigoACC (“Yesterday, Mary was reading this book”) is the 
distribution of case, in the sense that the internal argument is assigned nominative case in the 
se-sentence while accusative case appears to be absent. Otherwise, the syntactic spine is the 
same in both sentences in the sense that either se or a referential DP such as Marija is merged 
as the specifier of a VoiceP. 
Schäfer’s (2017, 5–6) account, however, posits that accusative case is actually present in the 
structure, and that it is assigned to se. Such a syntactic state of affairs arises due the fact that se
is taken to be a minimal pronoun that is completely unspecified for φ-features, in contrast with 
a fully-fledged referential DP such as Marija in the ordinary transitive variant. Since Schäfer 
(ibid.) takes a set of valued and interpretable φ-features to be a prerequisite for a DP to be 
assigned nominative case, the probe that triggers syntactic agreement skips se, which lacks 
such features, and assigns nominative to the other referential DP; that is, knjiga. By contrast, 
Schäfer takes accusative case to be a type of dependent case that is assigned to a DP “if a 
different DP has valued the accessible phase head via AGREE” (2017, 6). This means that
since the DP knjiga has been assigned nominative case by the regular process of syntactic 
agreement, se surfaces with dependent accusative case.
I take it that such an approach to the positioning of se and the consequent inverted distribution
of structural case is conceptually on the right track for Slovenian se-sentences, since it captures 
two very important and quite general structural facts about the construction. On the one hand, 
it predicts why, in terms of the morphological marking on the main verb, se-sentences pattern 
with ordinary active sentences. This is shown by the fact that a se-sentence realises the main 
verb as an l-participle in a past tense construction, as in (38). The reverse verbal morphology
holds of canonical Slovenian passives, where the main verb is an n-participle complementing 
the passive auxiliary biti (“be”) in the past tense, as seen in the sentence KnjigaNOM je bila
brana, which is the closest morphosyntactic equivalent of the English sentence The book was 
being read. According to Embick (2004, 150), a verb surfaces with passive morphology only 
if it lacks an overt external argument. A se-sentence, however, does contain such an overt
external argument, namely, se, hence its active morphology. On the other hand, this approach 
provides a straightforward explanation as to why the reflexive morpheme surfaces as se, and 
not for instance as dative si.  That is, the morphophonological form is such precisely because 
se carries (dependent) accusative case.
The core syntactic idea underlying (40) is that a sentence with se is built up in the same manner 
as a prototypical active transitive construction in terms of argument structure, in the sense that 
it contains both an internal argument in the vP , knjiga, as well as an external argument in the 
specifier of v oiceP; that is,  se. At first look, the only thing that distinguishes a se-sentence like (38), 
Ta knjiga
nom
 se je včeraj brala, from a prototypical active transitive sentence such as Marija
nom
 je 
včeraj brala to knjigo
acc
 (“Yesterday, Mary was reading this book”) is the distribution of case, in the 
sense that the internal argument is assigned nominative case in the se-sentence while accusative case 
appears to be absent. Otherwise, the syntactic spine is the same in both sentences in the sense that 
either se or a referential DP such as Marija is merged as the specifier of a v oiceP . 
Schäfer’s (2017, 5–6) account, however, posits that accusative case is actually present in the structure, 
and that it is assigned to se. Such a syntactic state of affairs arises due the fact that se is taken to be 
a minimal pronoun that is completely unspecified for φ -features, in contrast with a fully-fledged 
referential DP such as Marija in the ordinary transitive variant. Since Schäfer (ibid.) takes a set of 
valued and interpretable φ -features to be a prerequisite for a DP to be assigned nominative case, the 
probe that triggers syntactic agreement skips se, which lacks such features, and assigns nominative 
to the other referential DP; that is, knjiga. By contrast, Schäfer takes accusative case to be a type of 
dependent case that is assigned to a DP “if a different DP has valued the accessible phase head via 
AGREE” (2017, 6). This means that since the DP knjiga has been assigned nominative case by the 
regular process of syntactic agreement, se surfaces with dependent accusative case.  
I take it that such an approach to the positioning of se and the consequent inverted distribution 
of structural case is conceptually on the right track for Slovenian se-sentences, since it captures 
two very important and quite general structural facts about the construction. On the one hand, 
it predicts why, in terms of the morphological marking on the main verb, se-sentences pattern 
with ordinary active sentences. This is shown by the fact that a se-sentence realises the main 
verb as an l -participle in a past tense construction, as in (38). The reverse verbal morphology 
holds of canonical Slovenian passives, where the main verb is an n -participle complementing 
the passive auxiliary biti (“be”) in the past tense, as seen in the sentence Knjiga
NOM
 je bila 
brana, which is the closest morphosyntactic equivalent of the English sentence The book was 
7
 Note that the syntactic derivation in (40) was originally proposed by Alexiadou, Anagnostopoulou, and Schäfer (2015) and 
Schäfer (2017) for the morphemes sich (in German) and se/si (in Romance languages).

57 LANGUAGE
being read. According to Embick (2004, 150), a verb surfaces with passive morphology only if 
it lacks an overt external argument. A se-sentence, however, does contain such an overt external 
argument, namely, se, hence its active morphology. On the other hand, this approach provides 
a straightforward explanation as to why the reflexive morpheme surfaces as se, and not for 
instance as dative si. That is, the morphophonological form is such precisely because se carries 
(dependent) accusative case.
4.2 The Semantics of se
The present contribution is tied to the semantics of se. First recall from Section 2.3 that the 
phonologically null MiddleOp is posited as quite a complex semantic operator – it introduces 
existential quantification over the agent variable as well as generic quantification over situations. 
This explains why an English middle sentence is ill-formed if it is not interpreted generically, 
as in (39). By contrast, the corresponding Slovenian se-sentence is perfectly well-formed with a 
non-generic (i.e., eventive) interpretation (38), so it seems that se is functionally a simpler but 
overt variant of MiddleOp that is not tied to a generic quantifier, as in (41). 
(41)  
4.2 The Semantics of se
The present contribution is tied to the semantics of se. First recall from Section 2.3 that the 
phonologically null MiddleOp is posited as quite a complex semantic operator – it introduces
existential quantification over the agent variable as well as generic quantification over 
situations. This explains why an English middle sentence is ill-formed if it is not interpreted 
generically, as in (39). By contrast, the corresponding Slovenian se-sentence is perfectly well-
formed with a non-generic (i.e., eventive) interpretation (38), so it seems that se is functionally
a simpler but overt variant of MiddleOp that is not tied to a generic quantifier, as in (41).
(41) ⟦se⟧ = λf ⟨e, ⟨s,t⟩⟩.λe ⟨s ⟩.∃x[f(x, e)]
The semantics in (41) precisely illustrate Grahek’s (2008, 52) otherwise quite informal
observation that se reduces the external argument by operating on a predicate’s argument grid 
in the syntax.
8
Concretely under the present proposal, such argument reduction simply means
that se targets a syntactic structure that contains an open agent variable introduced by Voice,
and maps its denotation onto a scope in which it is existentially bound. (In terms of the θ-
Criterion, this means that se saturates the external θ-role in semantics and thus precludes its 
assignment to an external argument DP in syntax.)
Evidence for (41) lies in the fact that Slovenian se-sentences block PPs denoting the external 
argument. While a canonical Slovenian passive allows its external argument to be expressed 
via an od-phrase (42a) or a s-strani (42b) phrase, both PPs are ungrammatical in se-sentences 
(43), although an implicit agent is present in the semantic representation of both constructions.
(42) a. Sin je  bil pohvaljen         od očeta.
son.NOM AUX BE praise.N-PTCPL by father
“The son was praised by his father”
b.  Ta zadeva je   bila     že         sprožena         s strani mojega predhodnika.
this issue.NOM AUX BE already raise.N-PTCPL on part my      predecessor
“This issue was already raised by my predecessor.”
(Examples in (42) taken from Grahek (2008))
(43) a. Hiša           se      je    gradila          (*od delavcev).
house.NOM REFL AUX build.L-PTCPL (by workers)
“The house was being built (*by workers.)”
b. Jagode         so se      jedle             (*s strani mojega sina).
strawberries.NOM AUX REFL eat.L-PTCPL   (on part   my       son)
“Strawberries were being eaten (*by my son).”
The phrases are inadmissible in the two se-sentences in (43) because se, just like the null 
operator in English middles, requires as its argument an unsaturated Voice projection of type 
                                                
8
In relation to the semantics of se, the current proposal is different from Schäfer’s (2016, 14), which posits that 
se does not make a semantic contribution and that the agent variable, which is introduced by a special type of 
Voice head that Schäfer calls “E-bound Voice” (ibid.), is existentially bound from the very start. However, this 
would mean that the syntactic structure does not contain an open agent variable, which on the present account is 
problematic for the fact that Slovenian se-sentences with causative predicates allow the with-phrase (example (7)), 
which is licensed if the agent is initially an open variable in the syntactic-semantic representation.
The semantics in (41) precisely illustrate Grahek’ s (2008, 52) otherwise quite informal observation 
that se reduces the external argument by operating on a predicate’s argument grid in the syntax.
8
 
concretely under the present proposal, such argument reduction simply means that se targets 
a syntactic structure that contains an open agent variable introduced by voice, and maps its 
denotation onto a scope in which it is existentially bound. (In terms of the θ-c riterion, this 
means that se saturates the external θ-role in semantics and thus precludes its assignment to an 
external argument DP in syntax.) 
Evidence for (41) lies in the fact that Slovenian se-sentences block PPs denoting the external 
argument. While a canonical Slovenian passive allows its external argument to be expressed via 
an od-phrase (42a) or a s-strani (42b) phrase, both PPs are ungrammatical in se-sentences (43), 
although an implicit agent is present in the semantic representation of both constructions.
(42) a. Sin            je  bil pohvaljen         od očeta.
       son.nom aux be praise.n-ptcpl by father
     “The son was praised by his father”
  b.  Ta zadeva          je   bila     že         sprožena         s strani mojega predhodnika.
      this issue.nom  aux be   already    raise.n-ptcpl      on part my     predecessor
     “This issue was already raised by my predecessor.”
 (Examples in (42) taken from Grahek (2008))
8
 In relation to the semantics of se, the current proposal is different from Schäfer’s (2016, 14), which posits that se does not 
make a semantic contribution and that the agent variable, which is introduced by a special type of v oice head that Schäfer 
calls “E-bound v oice” (ibid.), is existentially bound from the very start. However, this would mean that the syntactic structure 
does not contain an open agent variable, which on the present account is problematic for the fact that Slovenian se-sentences 
with causative predicates allow the with-phrase (example (7)), which is licensed if the agent is initially an open variable in the 
syntactic-semantic representation.

58 Jakob Lenardič  The Middle Construction in Minimalist Syntax: English and Slovenian
(43) a. Hiša           se      je    gradila          (*od delavcev).
     house.nom refl aux    build.l-ptcpl (by workers)
     “The house was being built (*by workers.)”
 b. Jagode                   so   se      jedle             (*s strani mojega sina).
     strawberries.nom aux refl eat.l-ptcpl     (on part   my      son)
     “Strawberries were being eaten (*by my son).”
The phrases are inadmissible in the two se-sentences in (43) because se, just like the null operator 
in English middles, requires as its argument an unsaturated voice projection of type ⟨e,⟨s,t⟩⟩ , 
whose open agent variable it closes off with an existential quantifier (41). An od-phrase (42a) 
and a s-strani phrase (42b), however, are semantically equivalent to the English by -phrase in that 
they introduce the external argument into the structure, thereby closing the open agent variable 
by supplying the denotation of their complement DPs. Hence, admitting such phrases into the 
structure of a se-sentence would turn the denotation of the voice projection into that of type 
⟨s,t⟩, and thus preclude the merger of se. 
In addition, note that only anticausative verbs that are reflexively marked, such as odpreti (“open”) 
in (44), admit the Slovenian instrumental s-phrase, whereas a verb like počiti (“burst”), which 
is not reflexively marked, disallows it (45). On the present account, the contrast arises because 
sentence (44) contains an open agent variable, which is responsible for licensing the instrumental 
phrase and which gets closed off by se. conversely, sentence (45), lacking se, has a semantically 
vacuous voice head, so the s-phrase is disallowed (consequently, the possible inference of an 
agent popping the balloon originates pragmatically (cf.  Schäfer and v ivanco 2016)).
(44) Vrata se       zlahka  odprejo (s ključem).
   doors REFL  easily    open     (with key).
“These doors open easily with a key.”
(45) Ta balon    (*se)     zlahka poči      (*s škarjami).
  this balloon refl      easily pops      (with scissors).
“This balloon pops easily (*with scissors)”
Let’s now return to the intra-linguistic properties of Slovenian se-sentences outlined at the 
beginning of this section. Under the present proposal, the generic se-sentence in (37), T a knjiga se 
bere zlahka, differs from the eventive sentence in (38), T a knjiga se je včeraj brala, only minimally. 
In both cases, se first existentially binds the open agent variable, resulting in the fact that the 
denotation of v oiceP (40) is a function in which only the event variable remains open, as in (46).
(46) 
⟨e, ⟨s,t ⟩⟩, whose open agent variable it closes off with an existential quantifier (41). An od-
phrase (42a) and a s-strani phrase (42b), however, are semantically equivalent to the English 
by-phrase in that they introduce the external argument into the structure, thereby closing the
open agent variable by supplying the denotation of their complement DPs. Hence, admitting 
such phrases into the structure of a se-sentence would turn the denotation of the Voice 
projection into that of type ⟨s,t ⟩, and thus preclude the merger of se.
In addition, note that only anticausative verbs that are reflexively marked, such as odpreti
(“open”) in (44), admit the Slovenian instrumental s-phrase, whereas a verb like počiti
(“burst”), which is not reflexively marked, disallows it (45). On the present account, the
contrast arises because sentence (44) contains an open agent variable, which is responsible for 
licensing the instrumental phrase and which gets closed off by se. Conversely, sentence (45),
lacking se, has a semantically vacuous Voice head, so the s-phrase is disallowed (consequently, 
the possible inference of an agent popping the balloon originates pragmatically (cf.  Schäfer 
and Vivanco 2016)).
(44) Vrata se     zlahka odprejo (s ključem).
doors REFL easily  open     (with key).
“These doors open easily with a key.”
(45) Ta balon    (*se)     zlahka poči      (*s škarjami).
this balloon REFL easily pops      (with scissors).
“This balloon pops easily (*with scissors)”
Let’s now return to the intra-linguistic properties of Slovenian se-sentences outlined at the 
beginning of this section. Under the present proposal, the generic se-sentence in (37), Ta knjiga 
se bere zlahka, differs from the eventive sentence in (38), Ta knjiga se je včeraj brala, only 
minimally. In both cases, se first existentially binds the open agent variable, resulting in the 
fact that the denotation of VoiceP (40) is a function in which only the event variable remains 
open, as in (46).
(46) ⟦VoiceP ⟧ =  λe ⟨s ⟩.∃x[AGENT(x, e) ∧ READ(e) ∧ PATIENT(the book, e)]
The semantic difference between sentences (37) and (38) arises because, in the former
sentence, a generic quantifier, which is available as an independent semantic operator in all 
languages and is prototypically triggered when the sentence refers to an unbounded set of 
events temporally located in the present domain (Chierchia 1995), is introduced to saturate the 
remaining open event variable.
9
Applied to (46), the generic quantifier brings about the by now 
well-known middle LF, paraphrased as Generally, when (an arbitrary) someone reads the 
book, she does so easily.
4.3 The Covert English Middle Operator and the Slovenian se
As a final note, let’s consider the fact that English middles admit fewer verb types than the 
corresponding Slovenian se-sentences. Perhaps most strikingly, an English middle sentence is
ill-formed with a stative verb (47), which is otherwise admissible in the corresponding 
Slovenian variant (48).
                                                
9
For the eventive variant (38), I assume that existential closure is applied to the remaining event variable at TP,
so the sentence is interpreted as There was an event (temporally included in the past time interval denoted by 
yesterday) in which someone was reading the book (Dowty 1982).
The semantic difference between sentences (37) and (38) arises because, in the former sentence, 
a generic quantifier, which is available as an independent semantic operator in all languages and 
is prototypically triggered when the sentence refers to an unbounded set of events temporally 
located in the present domain (chierchia 1995), is introduced to saturate the remaining open 

59 LANGUAGE
event variable.
9
 Applied to (46), the generic quantifier brings about the by now well-known 
middle LF , paraphrased as Generally, when (an arbitrary) someone reads the book, she does so easily. 
4.3 The covert English Middle Operator and the Slovenian se
As a final note, let’s consider the fact that English middles admit fewer verb types than the 
corresponding Slovenian se-sentences. Perhaps most strikingly, an English middle sentence is 
ill-formed with a stative verb (47), which is otherwise admissible in the corresponding Slovenian 
variant (48).
(47) *This castle sees easily from afar.
      Intended: “Anyone can see this castle with ease from afar.”
(48)   Ta   grad            se     zlahka vidi od daleč.
     this castle.nom  refl   easily  see  from afar
    “Anyone can see this castle with ease from afar.”
The fact that (48) is a well-formed sentence should come as no surprise on the present account, 
as its grammaticality simply follows from the general sentence composition rules that have been 
proposed for Slovenian se-sentences.  concretely, the vP videti grad “see castle” merges with a v oice 
head introducing the external participant variable (in this case, the experiencer), which is then 
saturated when se is subsequently introduced as the specifier of v oiceP . The reason as to why (47) is 
ill-formed, however, is quite mysterious, especially given that MiddleOp is posited as an operator 
that simply targets a syntactic structure with an open variable denoting the external participant.
I am unable to offer a precise explanation of this fact, but conclude on a more informal and 
speculative note by tentatively proposing that the difference in grammaticality between (47) and 
(48) is due to the fact that the English lexicon lacks a lexical item such as se, which is a nominal 
pronoun. In other words, English middles are formed by an operator that is not lexicalised as an 
independent nominal expression, while the operator itself plays a more complex semantic role 
than se, invariably introducing generic quantification over the event variable. In short, because 
MiddleOp in English middles is a more complex operator than se in Slovenian sentences, it is 
also more limited in its application.
10
  
9
 For the eventive variant (38), I assume that existential closure is applied to the remaining event variable at tP , so the sentence 
is interpreted as There was an event (temporally included in the past time interval denoted by yesterday) in which someone was 
reading the book (Dowty 1982).
10
 In this respect, an anonymous reviewer questions why MiddleOp is, on the present account, taken to be a different syntactic 
constituent (i.e., the head of its own phrase above v oiceP) from the reflexive morpheme se, which is posited as the specifier 
of v oiceP . The reviewer claims that such an analysis potentially misses a cross-linguistic generalisation about the grammatical 
properties that English middles and Slovenian se-sentences have in common. I would like to emphasise that this is precisely the 
point – the English middle construction and the Slovenian se-construction are syntactically different beasts. English middles, 
lacking a nominal element like se, are structurally deficient, which is why their well-formedness seems to be dependent on 
genericity (according to Härtl (2012), genericity obviates the requirement that a transitive verb like read combine with an 
external argument, which is shown by the fact that an eventive sentence like *This book read yesterday is bad precisely because 
no such argument is present in its structure) and why they cannot be formed from just about any predicate type (example 
(47)), whereas the formation of  se-sentences is unconstrained (example (48)). In other words, the present proposal highlights 
the fact that the equivalent interpretation of Slovenian generic se-sentences and English middles arises due to different – albeit 
– similar underlying operations (i.e., MiddleOp binds the agent and event variables in one fell swoop, while se only binds 
the agent variable). Furthermore, note that this is consistent with Lekakou’s (2004) cross-linguistic observation that different 
languages may not employ the same grammatical strategies to form middles.

60 Jakob Lenardič  The Middle Construction in Minimalist Syntax: English and Slovenian
5 Conclusion
This paper has discussed the formation of English and Slovenian middle sentences from the 
perspective of a minimalist syntactic approach to argument structure. The paper has first 
presented the conceptual and empirical advantages of Bruening’s proposal (2012), which posits 
that English middles are derived via a null operator that requires an unsaturated v oice projection, 
showing how his approach elegantly (i.e., without additional stipulations) accounts for certain 
key properties of the construction, such as the arbitrary reference of the agent. In addition, it has 
been proposed that the adverb is not a semantic argument of the null middle operator. Lastly, 
the paper has discussed the reflexive morpheme se in Slovenian se-sentences, and proposed that 
it has a semantics similar to that of the null English operator.
References
Ackema, Peter, and Maaike Schoorlemmer. 1994. “The Middle construction and the Syntax-Semantics 
Interface.” Lingua 93 (1): 59–90. https://doi.org/10.1016/0024-3841(94)90353-0. 
Alexiadou, Artemis. 2010. “On the Morphosyntax of (Anti)-causative v erbs.” In Lexical Semantics, Syntax, 
and Event Structure, edited by Malka Rappaport Hovav, Edit Doron, and Ivy Sichel, 177–203. Oxford: 
Oxford University Press. https://doi.org/10.1093/acprof:oso/9780199544325.003.0009. 
Alexiadou, Artemis, Elena Anagnostopoulou, and Florian Schäfer. 2015. External Arguments in Transitivity 
Alternations: A Layering Approach. Oxford: Oxford University Press.
Borer, Hagit. 2004. “The Grammar Machine.” In The Unaccusativity Puzzle, edited by Artemis Alexiadou, 
Elena Anagnostopoulou, and Martin Everaert, 288–331. Oxford: Oxford University Press. https://doi.
org/10.1093/acprof:oso/9780199257652.003.0012.
Bruening, Benjamin. 2012. “By Phrases in Passives and Nominals.” Syntax 16 (1): 1–41. https://doi.
org/10.1111/j.1467-9612.2012.00171.x.
chierchia, Gennaro. 1995. Dynamics of Meaning. chicago: University of chicago Press.
chomsky, Noam. 2001. “Derivation by Phase.” In Ken Hale: A Life in Language, edited by Michael 
Kenstowicz, 1–52. cambridge, MA: MIt Press.
Dowty, Don. 1982. “ tense, t ime Adverbs and compositional Semantic Theory.” Linguistics and Philosophy 5: 
23–55.
Embick, David. 2004. “Unaccusative Syntax and v erbal Alternations.” In The Unaccusativity Puzzle, edited by 
Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert, 137–58. Oxford: Oxford University 
Press. http://dx.doi.org/10.1093/acprof:oso/9780199257652.003.0006.
Fagan, Sarah M. B. 1992. The Syntax and Semantics of Middle Constructions: A Study with Special Reference to 
German. cambridge: cambridge University Press.
Fellbaum, christiane. 1986. On the Middle Construction in English. Bloomington, Indiana: Indiana University 
Linguistics club.
Grahek, Sabina. 2008. “Middles in Slovene.” Leeds Working Papers in Linguistics and Phonetics 13: 44 –75. 
Härtl, Holden. 2012. “Generic Rescue: Argument Alternations and the Monotonicity condition.” In Repairs: 
The Added Value of Being Wrong, edited by Patrick Brandt and Eric Fuß, 95–130. Berlin: De Gruyter. 
Klingvall, Eva. 2015. “A Secret Agent in the Middle?” Working Papers in Linguistics (Department of English 
in Lund) 5: 91–114. Accessed August 23, 2018. http://www.sol.lu.se/fileadmin/media/forskning/
workingpapers/engelska/vol05/klingvall-wp05.pdf. 

61 LANGUAGE
Kratzer, Angelika. 1996. “Severing the External Argument from Its v erb.” In Phrase Structure and the Lexicon, 
edited by Johan Rooryck and Laurie Zaring, 109–37. Dordrecht: Kluwer Academic Publishers. https://
doi.org/10.1007/978-94-015-8617-7_5.
Krifka, Manfred, Francis Jeffry Pelletier, Gregory N. carlson, Alice ter Meulen, Godehard Link, and Gennaro 
chierchia. 1995. “Genericity: An Introduction.” In The Generic Book, edited by Gregory N. carlson and 
Francis Jeffry Pelletier, 1–124. chicago: University of chicago Press.
Legate, Julie Anne. 2014. Voice and v – Lessons from Acehnese. Massachusetts, MA: MIt Press.
Lekakou, Marika. 2004. “Middles as Disposition Ascriptions.” In Arbeitspapier Nr. 117: Proceedings of the 
Conference “Sub8 – Sinn und Bedeutung”, edited by cécile Meier and Matthias Weisgerber, 181–196. 
Konstanz: Universität Konstanz.
Levin, Beth, and Malka Rappaport Hovav. 1995. Unaccusativity: At the Syntax-Lexical Semantics Interface. 
cambridge, MA: MIt Press.
Marantz, Alec. 1997. “No Escape from Syntax.” University of Pennsylvania Working Papers in Linguistics 4: 
201–25.
Marelj, Marijana. 2004. Middles and Argument Structure Across Languages. Utrecht: LO t Publishing. 
Mcconnell-Ginet, Sally. 1994. “On the Non-Optionality of certain Modifiers.” In Proceedings of SALT 4, 
edited by Mark Harvey and Lynn Santelmann, 230–50. https://doi.org/10.3765/salt.v4i0.2458. 
Ramchand, Gillian. 2008. Verb Meaning and the Lexicon: A First Phase Syntax. cambridge: cambridge 
University Press.
Rivero, Maria-Luisa, and Milena Milojević Sheppard. 2003. “Indefinite Reflexive clitics in 
Slavic: Polish and Slovenian.” Natural Language & Linguistic Theory 21: 89–155. https://doi.
org/10.1023/A:1021841517604. 
Schäfer, Florian. 2007. “Middles as v oiced Anticausatives.” In Proceedings of NELS 37 (Volume 2), edited 
by Emily Elfner and Martin Walkow, 183–96. Accessed August 31, 2018. http://ling.auf.net/
lingbuzz/000400. 
—. 2016. “Medio-Passives within a Formal typology of v oice.” Manuscript. Accessed July 20, 2018. https://
amor.cms.hu-berlin.de/~schaeffl/papers/mediopassives2.pdf. 
—. 2017. “Romance and Greek Medio-Passives and the typology of v oice.” In The Verbal Domain, edited by 
Roberta D’Alessandro, Irene Franco, and Ángel J. Gallego, 129–52. Oxford: Oxford University Press. 
https://doi.org/10.1093/oso/9780198767886.003.0006.
Schäfer, Florian, and Margot v ivanco. 2016. “Anticausatives are Weak Scalar Expressions, Not Reflexive 
Expressions.” Glossa: A Journal of General Linguistics 1 (1): 1–36. http://doi.org/10.5334/gjgl.36.