HELLENISTIC GREEK MIDDLE VOICE:
SEMANTIC EVENT STRUCTURE AND VOICE TYPOLOGY
by
RACHEL E. AUBREY

Bachelor of Arts in Literature and Latin, Calvin University, 2006
Certificate of Applied Linguistics, Graduate Institute of Applied Linguistics, 2008

Thesis
Submitted in Partial Fulfillment of the Requirements for
the Degree of

MASTER OF ARTS IN LINGUISTICS AND TRANSLATION
in the
FACULTY OF GRADUATE STUDIES
TRINITY WESTERN UNIVERSITY

March 2020
© Rachel E. Aubrey, 2020

i
Abstract
This thesis advocates a semantic approach to Hellenistic Greek middle voice, endeavoring to
capture a variety of middle expressions and their internal semantic relations. Various event types
that receive middle expression in Greek form a continuum; they adopt the scale of semantic
transitivity as a conceptual foundation for middle phenomena, among middle systems crosslinguistically and in Ancient Greek (Kemmer 1993). Historical traditions in voice analysis point
to syntactic relationships, with alternations framed as choices in clausal subject. Such narrow
definitions do not capture the semantic behavior of the Greek middle. Neglecting differences in
semantic event structure overlooks fundamental aspects of the Greek voice system. The present
analysis describes Greek voice in terms of meaning-oriented distinctions in event structure, as
they pertain to shifts in both the type of action and attentional focus regarding facets of an event
frame, engaging semantic and pragmatic motivations in voice (Langacker 2006, Shibatani 2006).

ii
Table of Contents
1. Introduction ............................................................................................................................. 1
1.1 Challenges of the middle ........................................................................................... 5
1.1.1 Functional diversity ......................................................................................... 5
1.1.2 Formal variation ............................................................................................ 10
1.2 Motivations in voice ................................................................................................ 15
1.3 Organization ............................................................................................................. 21
2. Middle typology .................................................................................................................... 23
2.1 Voice systems: derived vs. basic ............................................................................. 24
2.2 Syntactic accounts and semantic regularities ........................................................... 31
2.2.1 Anticausatives and passives .......................................................................... 32
2.2.2 Consistency in the lexicon............................................................................. 39
2.3 Inventory of middle types ........................................................................................ 49
3. Middle diachrony .................................................................................................................. 56
3.1 Lexicalization in voice ............................................................................................. 57
3.1.1 Formal idiosyncrasies .................................................................................... 58
3.1.2 Semantic idiosyncrasies ................................................................................ 61
3.2 Grammaticalization of passive morphology ............................................................ 64
3.2.1 -(θ)η- development......................................................................................... 66
3.2.2 -(θ)η- expansion ............................................................................................. 73
4. Event construal and the nature of voice ................................................................................ 78
4.1 Semantic transitivity ................................................................................................ 80
4.2 Adjustments in energy and attention........................................................................ 82
4.2.1 Alternations in voice: Shift in primary focus ................................................ 85
4.3 Alternations in voice: External vs. internal energy .................................................. 89
4.3.1 Change of state: Physical process ................................................................. 89
4.3.2 Change of state: Body motion ....................................................................... 94
4.3.3 Change of state: Collective motion ............................................................... 98
4.3.4 Change of state: Mental process.................................................................. 100
4.4 Alternations in voice: Symmetrical energy in cognitive events ............................ 105
4.4.1 Mental activity............................................................................................. 107
4.4.2 Speech act .................................................................................................... 112
4.4.3 Perception .................................................................................................... 116
4.5 Alternations in voice: Symmetrical energy in cyclic events .................................. 121
4.5.1 Reciprocal.................................................................................................... 122
4.5.2 Grooming/Direct reflexive .......................................................................... 127
4.5.3 Self-benefactive/Indirect reflexive .............................................................. 134
4.6 Synthesis ................................................................................................................ 139
5. Conclusion .......................................................................................................................... 142
References ................................................................................................................................... 146

1

1. Introduction
One of the primary challenges of describing Ancient Greek middle voice is its multifunctional
character. Its semantic diversity (e.g. reflexive, reciprocal, benefactive, anticausative, passive,
etc.) renders it difficult to summarize with simple generalizations (Kemmer 1993, Rijksbaron
2002, Shibatani 2004). Standard descriptions in New Testament (NT) Greek grammars offer
limited help in this regard, as they are largely rooted in classical treatments of voice. Primacy is
given to the active-passive contrast, with almost exclusive focus on morphosyntax as a
descriptive framework. This strategy may seem sufficient at first, especially with the activepassive distinction and its close alignment of syntax and semantic roles. But its application to the
middle exposes its deficiencies as an explanatory tool. The lack of a distinct syntactic realization
or singular semantic role to associate with the middle brings to light its limitations.
Overreliance on morphosyntax in the description of voice has consequences for our
understanding of the Greek middle. First, it sidesteps important developments in typological
research that point to the polysemous nature of middle voice phenomena and their functional
similarity across a variety of genetically non-related languages.1 Second, traditional and
contemporary accounts of Greek middle voice demonstrate a tendency toward overly simplified
descriptions of middle semantics. By giving primacy to the active-passive alternation and its
morphosyntactic expression, they are left with accounts of the middle that swing in wildly
divergent directions. They either (1) list its usages as discrete and otherwise unrelated categories
with all of the variety and none of the coherence.2 Or (2), they focus on necessarily simple
1

Representative languages include ancient languages, e.g. Greek (Allan 2003) and Hebrew (Wolde 2019),
as well as modern Indo-European (IE), e.g. Spanish, French, German (Kemmer 1993) and non-IE languages, e.g.
Somali [Cushitic, Somalia] (Saeed 1995) and Halkomelem [Salish, British Columbia] (Gerdts and Hukari 2006).
2
See Gildersleeve 1900, 64-79; Wallace 1996, 414; Rijksbaron 2006, 161-69; Mounce 2009, 228-29. The
challenge is that delineated types differ with each description, making it hard to parse one from another. Ladewig

2

generalizations, e.g. ‘subject affectedness’ or ‘participant involvement’.3 Each approach, in its
own way, construes voice functions too narrowly (Kemmer 1993, 4).4
The end result is that approaches in NT Greek grammar fail to understand middle voice
in its typological context, on the one hand, and its paradigmatic context on the other. I propose
an alternative model that (1) draws on typological regularities among middle voice systems, (2)
contrasts middle usage patterns relative to their active counterparts, and (3) frames voice
alternations5 in terms of the larger forces of semantic transitivity.
My claim is that semantic transitivity, as a fundamentally scalar notion, provides the
conceptual foundations for Greek middle voice, with departures from the basic transitive event
allowing for a variety of semantic contrasts in event structure. In this way, transitivity affords a
cognitive frame for understanding a variety of voice alternations, all of which pertain to various
aspects of event development, or how actions unfold (Shibatani 2006). With this conceptual
grounding, I advance a semantic approach to the Greek middle that not only seeks to differentiate
middle uses according to semantic event types, but also aims to clarify how they relate to one
another as well as how they function within an integrated system of cognitively related
categories involving voice and transitivity (Kemmer 1993, Langacker 2006).
To gain a better understanding of the semantic range of middle voice phenomena across
languages and in Ancient Greek, I review current approaches to the Greek middle (chapter 1) and

(2010, 106) restricts the middle to direct and indirect reflexive uses. Mathewson and Emig (2016, 148-51) include
reflexive, intensive, reciprocal, and intransitive middles. Wallace (1996, 416-30) lists reflexive, redundant,
causative, permissive, reciprocal, and deponent middles. Often the enumeration of middle types relies more on
translational convenience than on grammatical distinctions in the Greek language.
3
See, for example, Matthewson and Emig (2019, 148-49).
4
Such generalizations define various middle uses with a single abstract feature. Popular characterizations
that simplify middle semantics then play an outsized role in describing the category; scholars and exegetes pick a
term, e.g. subject affectedness, and use it as a defining feature for why any given event occurs in middle voice.
5
i.e. verbs that inflect for active and non-active forms

3

then illustrate their shortcomings within a larger typological context (chapter 2). Particular
emphasis is given to: (a) the fundamental differences that exist between middle voice systems
and active-passive systems, (b) the tight parallels that exist among middle systems, such as
semantic restrictions that occur among middle alternations in regard to semantic classes of verbs
that permit one type of alternation or another, and (c) the recurring lexical event types that form
the semantic locus for middle expressions across languages.
Middle typology in chapter 2 gives rise to middle diachrony in chapter 3. I examine
processes of language change that affect the Hellenistic Greek voice system, specifically the
natural idiosyncrasies that arise within lexical semantics (§3.1) and the grammaticalization of the
-(θ)η- voice form (§3.2). The -(θ)η- form has historically been used as evidence that Greek
sustains a unique marker of passive voice apart from either the active or middle. Analyzing the
origins of -(θ)η- in early Greek (via Proto-Indo-European [PIE]) and tracing its diachronic shifts
through Classical and Hellenistic periods helps to (a) illustrate its functional diversity (it is used
for functions beyond the passive), (b) trace its functional parallels with other middle-passive
forms, and (c) map out the semantic network of middle types and how they relate to one another.
Finally, I present an integrated and unified account of Greek middle voice (chapter 4),
building on insights from typology (chapter 2) and processes of language change (chapter 3) in
order to describe the Greek voice system in a way that accounts for its diversity of usage, while
also providing a motivated account of its structure and unity of function.
Application of language typology and diachronic research provides practical tools for
addressing key difficulties in the description of the Greek middle. First, it includes a robust and
productive replacement for the problem of deponency in NT Greek studies. Voice typology helps
to establish what constitutes canonical norms among middle voice systems, especially in regard

4

to the distribution of middle morphology across the verbal lexicon. This kind of insight offers an
appealing alternative to relying on English translation value for understanding Greek middle
forms, a practice common in NT Greek studies. Second, it includes insights into formal and
semantic idiosyncrasies that regularly arise in Greek middle voice (and indeed across middle
systems generally) but are not common to active-passive systems. With the backdrop of language
typology, what appears to be language-specific anomaly actually reflects larger semantic patterns
that exist across middle voice systems. These are patterns that are often overlooked and left
unaccounted for within the classical active-passive framework for voice.
Beyond NT Greek studies, there are also benefits to crosslinguistic work in translation
and language description. Providing a broader understanding of the Greek voice system and the
kinds of semantic alternations it evinces helps in bridging the gap between source language and
target language translation. Literature in voice and language typology also benefit. This is true in
a broad sense as well as in a more specific sense within Greek language studies.
A semantic account of the Ancient Greek middle contributes to the growing body of
literature in voice analysis. In doing so, it furnishes an empirical basis for building crosslinguistic generalizations as well as a testing ground for refining typological claims, as in
Geniušienė (1987), Klaiman (1991), and Kemmer (1993). It is perhaps Kemmer (1993) who
represents a deliberate turn toward cognitive approaches to voice, with an interest in how
language categories reflect cognitive categorization and our ability to construe events in different
ways. My account of the Greek middle offers a language-specific application of broader crosslinguistic claims from cognitively-based approaches to grammatical voice, as in Kemmer (1993),
Langacker (2006), Shibatani (2006), Maldonado (2007), and Næss (2007).

5

In regard to Greek language studies, my account of the middle in the Hellenistic period
contributes to diachronic investigations of the Greek voice system. Rutger Allan (2003) supplies
a semantic analysis of the Greek middle in the Classical period, drawing from Classical texts for
analysis. Linda Manney (2000) does much the same, but for Modern Greek usage patterns. My
intention is to continue these discussions, by adding data and observations from Hellenistic
Greek texts (e.g. Philo, Josephus, Septuagint, New Testament texts, Apostolic Fathers, etc.),
providing fodder for research in language change and how middle systems evolve over time.
1.1 Challenges of the middle
A number of issues arise in the description of the Greek middle: (1) its semantic basis is unclear.
Voice categories in Greek and across languages are multifunctional and this makes it difficult to
draw boundaries around what is or is not ‘middle’ (Kemmer 1993, 1-2; Shibatani 2004); (2) its
formal expression is uncertain, particularly due to variation in Greek morphology and issues in
how voice morphology relates to syntactic structure (Klaiman 1991, 1-43; Allan 2003, 126-202).
1.1.1

Functional diversity

First, the diversity of middle types makes it difficult to establish what constitutes middle
functions (Robertson 2006 [1919], 803; Wackernagel 2009, 164). It is frequently treated in
relation to reflexives (though this is a misrepresentation of the functions it serves) across
languages and in Ancient Greek (Allan 2003).6 Yet identifying functions that go beyond standard

6

This elusiveness is partly due to cross-linguistic variety in voice. A voice category in one language may
subsume several different expressions in another (Geniušienė 1987; Shibatani 2004, 1157). The Greek middle form
includes multiple functions that are distinct constructions in other languages, e.g. reflexives, reciprocals, cognition,
passives, and anticausatives (Klaiman 1991, 44-109; Kemmer 1993, 16-20). Even if two languages have similar
expressions with some functional overlap, there is rarely if ever exact equivalence at the level of morphology and
syntax (Haspelmath 2003). This is true for English reflexives and the Greek middle. The English reflexive pronoun
is a popular placeholder for glossing Greek middle verbs, but it hardly means they are coterminous. Consider middle
types that resemble reflexive actions, e.g. grooming (κατακαλύπτομαι ‘cover oneself’) and motion (τρέπομαι ‘turn
(oneself)’). In both, middle morphology shows some semantic affinity for the use of the English reflexive. But the

6

examples (λούμαι ‘wash oneself’), only makes the middle more difficult to define, especially
given how voice is typically conceived in the Greek grammatical tradition (Geniušienė 1987, 8).7
Traditionally, voice is described as a relationship between subject and verb. A change in
voice morphology correlates with a shift in the semantic role of the subject in relation to the verb
in syntactic structure (Wallace 1996, 408-9; Mathewson and Emig 2016, 142). The subject of the
active does the action, as agent. The subject of the passive suffers the effects of the action, as
patient (Lyons 1968, 372-73; Mounce 2009, 228). But the middle shows more variety than
syntactic rules can predict and does not reliably correlate with a distinct semantic role for the
subject. Because of this, it is often described by analogy: The subject of the middle does the
action (like the active) but also receives the action (like the passive) (Klaiman 1991, 3).
In this overly narrow conception of voice, the middle is often treated as though it is
semantically equivalent to the Greek reflexive. In a typical reflexive, a reflexive form ἑαυτόν is
used to mark coreference of participants in situations in which the two are normally distinct. In a
transitive event like (1.1), there are two distinct participants (agent, patient) and two distinct
entities (king, servant). In a reflexive expression in (1.2), two participants (agent, patient) are
evoked but instead of referring to two distinct entities, they are filled by just one referent. The
reflexive form ἑαυτόν is used with transitive (two-participant) events to signal the non-default
case that agent and patient are coreferential. It provides a means of overtly expressing that an
agent acts on himself rather than on another possible entity (Faltz 1977; Kemmer 1993, 42-44).

same relation does not hold for other middles: ἀποτάσσομαι ‘bid farewell’, μίγνυμαι ‘mingle, copulate’, διαθήσομαι
‘agree, make a covenant’, πείθομαι ‘obey, believe’, γεύομαι ‘taste’, and ὀργίζομαι ‘become angry’.
7
By traditional conceptions in Greek grammar, I mainly refer to pedagogical treatments for classroom
instruction, e.g. Young (1994), Wallace (1996), Mounce (2009), Mathewson and Emig (2016). Though, this
tradition stems from older works, e.g. Gildersleeve (1900), Robertson 2006 [1919], Dana and Mantey (1927), and
continues in more recent accounts, e.g. Lavidas and Papangeli (2007) and Ladewig (2010).

7

(1.1)

ἔβαλεν ὁ βασιλεὺς αὐτὸν εἰς φυλακὴν
The king threw him [the servant] into prison (Matthew 18:30)

(1.2)

Σίμων Πέτρος ἔβαλεν ἑαυτὸν εἰς τὴν θάλασσαν
Simon Peter threw himself into the sea (John 21:7)
The problem with treating middle morphology as semantically equivalent to the Greek

reflexive is that it ignores differences in semantic event types and creates a spurious impression
of the semantic range of middle expressions. This reliance on the reflexive analogy in describing
the Greek middle often begins with Greek grammars citing instances of the middle that involve a
similar kind of coreference, as with grooming verbs in (1.3) (Wallace 1996, 416-17). A volitional
subject instigates an action that affects a change directly on one’s own body.
(1.3)

θερμαίνομαι ‘heat/warm oneself, get warm’
νίπτομαι ‘wash (oneself)’
It is tempting to treat such cases as equivalent to the use of the Greek reflexive pronoun.

In (1.3) and (1.4), middle (-μαι) and reflexive (ἑαυτόν) forms are used to denote actions that are
carried out on oneself. A single entity is both starting point and endpoint of his own action.
(1.4)

ἀγαπᾷ ἑαυτὸν ‘he loves himself’
περιέπειραν ἑαυτοὺς ‘they stabbed themselves’
Yet, while middle and reflexive forms do show some functional overlap, they also differ

in a few key aspects. First, as in (1.3), the middle form consistently occurs with semantic events
that are normally performed by human referents on the human body. The middle specifies a set
of socio-cultural practices that are customarily carried out by human agents on themselves. The
reflexive form, however, occurs with a much wider class of verbs, both active and middle, that
represent actions that are not conventionally reflexive and are not typically performed on oneself,
e.g. σωσάτω ἑαυτὸν ‘he should save himself’, ἀρνέομαι ἐμαυτοῦ ‘I deny myself’.8 This formal

8
The reflexive form can be used to express regular grooming actions, particularly in contrastive contexts,
e.g. John 21:18. Active ζώννυμι ‘gird’ is used with the reflexive form to mark an explicit contrast between when

8

difference in Greek reflects a conceptual distinction between typical body actions and a much
larger class of events that are ordinarily performed on a distinct entity (Kemmer 1993, 58).
Second, in Greek as well as in other languages with middle voice (see Ch. 2), typical
body actions (e.g. grooming events in (1.3)), regularly receive middle expression. Their formal
marking pattern (middle morphology) does not pair them with actions that are typically carried
out on a distinct participant (e.g. ἀγαπάω ‘love’ in (1.4)). These actions require a reflexive
pronoun in order to express reflexive action (an agent acts on itself). Instead, grooming events
are formally expressed in the same way as other situation types that also involve fusion between
starting point and endpoint of the action. These events specify a range of actions, from those that
are less prototypically reflexive (1.5) to those that are semantically non-reflexive (1.6)-(1.8)
(Faltz 1977, 7; Kemmer 1993, 53).
One type that is often subsumed in reflexive analyses of the middle is the indirect
reflexive or benefactive type (see Ladewig 2010). Rather than a direct action on the body, verbs
in (1.5) involve a second figure. An agent acts on a distinct patient but does so as a recipient or
beneficiary of the action. The agent and beneficiary are coreferential.
(1.5)

περιποιέομαι ‘acquire something, gain possession of’9
προσλαμβάνομαι ‘take aside, bring along’

you, as a young man, clothe yourself (ἐζώννυες σεαυτὸν) as opposed to when you, as an old man, will submit to
others clothing you (ζώσει σε). The middle form of the same verb is used when there is no explicit contrast between
actions performed on oneself vs. those performed on others. In Acts 12:8, the context provides a situation in which it
is normal and expected to put on one’s own clothes, “Clothe yourself (Ζῶσαι) and put on your sandals”.
9
To clarify, when a reflexive pronoun occurs with a middle verb like περιποιέομαι (e.g. 1 Tim. 3:12-13), it
is not serving as an additional marker of middle semantics as is often assumed, but its presence in the clause serves
its own particular function. Often, this is contrastive stress, especially in contexts of possible ambiguity in which
there is more than one activated referent. The reflexive form explicitly singles out that the action occurs in regard to
a single participant to the exclusion of other possible discourse referents: ‘Deacons should manage their children and
households well…those who serve well gain good standing for themselves (περιποιοῦνται ἑαυτοῖς)’. The reflexive
form disambiguates the referent. It stipulates that it is the deacons, specifically those who serve well, who gain good
standing for themselves, not for other potential referents, e.g. other deacons, children, or household members.

9

Other middle types are less amenable to a strictly reflexive notion of the middle form.
Those in (1.6) do not express actions carried out on oneself, but reciprocal action (a), sensory
perception (b), mental activity (c), and speech (d). An agent acts volitionally, but like those in
(1.5), they typically engage a second participant in the action: fight s.o., touch s.o./s.t., etc.
(1.6)

(a) μάχομαι ‘fight with, contend’ (c) ἐκκρέμαμαι ‘pay attention to’
(b) ἅπτομαι ‘touch’
(d) ἀμείβομαι ‘answer, reply’
Middles in (1.7)-(1.8) are different still. For passive expressions in (1.7), the subject does

not instigate the event but only undergoes an action brought about by a second figure. In (a), the
subject is an experiencer – a sentient participant who receives a sensory impression, usually with
verbs of cognition, emotion, or perception. In a passive, the subject experiences a mental state
brought about by another participant. In (b), the subject is closer to a patient that undergoes a
physical effect or change of state/location (Langacker 1991, 285; Kemmer 1993, 128-29, 147ff.).
(1.7)

(a) παρακαλοῦμαι ‘be consoled (by)’ (b) βάλλομαι ‘be thrown (by)’ 10
For middle verbs in (1.8), the subject undergoes some kind of change, so it is patient-like,

but no second figure brings about the event. Instead, the subject does the action (like (1.6)), but it
is less agentive and does not volitionally instigate the event in the same way as (1.3) and (1.5).
(1.8)

φοβέομαι ‘fear, feel afraid’
συνάγομαι ‘gather (of a group)’

ἀπόλλῠμαι ‘perish, die’
καταποντίζομαι ‘drown/sink’

The same morphology expresses reflexive action and the passive function, along with
events in between that fit neither description. It is a trend to cite Lyons (1969, 373) in this regard:
In the middle, “the ‘action’ or ‘state’ affects the subject of the verb or his interests”, but Lyons is
largely echoing those who have gone before (Goodwin 1895, 256-58; Gildersleeve 1900, 64-79;

10

In a passive construction with βάλλομαι the subject may be animate or inanimate: ‘If a competitor…is
caught cheating, he is flogged, disqualified, and (ἔξω βάλλεται τοῦ σταδίου) thrown out of the stadium’ (II Clement
7.4) vs. ‘Every tree that does not bear good fruit is cut down and (εἰς πῦρ βάλλεται) thrown into the fire’ (Matt. 7:19).

10

Smyth 1956, 389-398). Other recent portrayals also follow traditional schemes, with the middle
adding a broad semantic nuance to the grammatical subject: self-interest, personal involvement,
participation, special focus, or subject-affectedness à la Lyons (Porter 1994, 63, 67; Wallace
1996, 414-15; Mounce 2009, 228; Decker 2014, 235-36; Mathewson and Emig 2016, 148). As
much as descriptions of this nature aim to find a reliable ‘middle quality’ that is abstract enough
to be equally pertinent to all middle event types regardless of their differences, they only suffice
as a starting point since they hardly capture the various functions the middle serves within the
voice system, let alone how they relate to one another, or the factors that motivate voice
alternations in the first place (Kemmer 1993, 1-4; Shibatani 2006, 217-18).
1.1.2

Formal variation

The second challenge is variation in the formal expression of the middle, particularly at the
intersection of morphology and syntax. For each aspect stem in Greek, there are separate verbal
affixes for active and middle-passive voices. All finite, and some nonfinite, verbs obligatorily
mark a voice form. From a morphosyntactic perspective, if ‘normal’ or ‘canonical’ verbs inflect
for each voice type, then all verbs are expected to do so, filling out verbal paradigm charts with
middle-passive forms and their active morphological counterparts (Baerman 2007, 2; Corbett
2007, 21-31, Ladewig 2010, 121-25). Yet systematic patterns like this in natural language are
always relative (Shibatani 2006, 219). Even in the midst of well-integrated patterns, there are
often gaps and variations that simply do not exist across all verbs (Aikhenvald 2007, 55).
Consider the presence of activa and media tantum. Activa tantum (active-only) verbs
constitute a set of intransitive one-participant verbs expressed only in active morphology. Verbs

11

like διδράσκω ‘walk, run’ and θρώσκω ‘jump’ do not inflect for middle-passive forms.11 Media
tantum (middle-only) verbs are expressed only in middle morphology and do not inflect for
active forms. Examples include transitive (two-participant) verbs (βούλομαι ‘want, prefer’, δέομαι
‘ask’, λογίζομαι ‘calculate’) and intransitive (one-participant) verbs (δύναμαι ‘be able’, ἅλλομαι
‘jump’, ἔρχομαι ‘go’) (Klaiman 1991, 98-99; Allan 2003, 49-52, 243-44). The presence of activa
and media tantum in Greek has been a long-standing source of discussion regarding (1) whether
such verbs ought to participate in voice alternations (i.e. inflect for active and non-active forms),
and (2) what their non-alternating status suggests about their meaning.12
A related area of confusion is the presence of a distinct -(θ)η- form in the perfective
paradigm (aorist/future). It is a matter of some debate because it varies from standard
expectations (Lavidas and Papangeli 2007, 102; Mathewson and Emig 2016, 152). As a voice

11

In addition, there are a number of Greek verbs that prefer middle morphology in the future but active
forms elsewhere, e.g. γινώσκω (present active) ~ γνώσομαι (future middle); εἰμί ~ ἔσομαι; ὁράω ~ ὄψομαι. See Emde
Boas et al. (2019, 191-93) for examples. Typological research suggests that languages with middle morphology have
a number of shared characteristics that warrant further investigation. One of them is the tight parallels that exist
among semantic classes that receive middle marking across languages. I discuss this in ch. 2. Another is the
interaction of middle marking with other verbal categories, e.g. tense, aspect, and modality. Klaiman (1991, 96)
notes that languages like Fula (Niger-Congo) and Ancient Greek both show some affinity or correlation between the
middle form and the future tense/modality. A subset of verbs in each language prefers the middle form when
expressing future action. This does not mean, however, that every middle-marking language shows the same
affinity; Tamil (Dravidian, South India) does not express the same correlation between future action and middle
forms (Klaiman 1991, 74). In regard to Greek, Willi (2018, 445-52) discusses the correlation between middle
morphology and future action as morphological matter, due to a shift that occured in early Greek. Before a distinct
future form had been created in early Greek, the sigmatic aorist subjunctive could be used to express future action.
Subsequently, the subjunctive stem *CeC-s-e/o- was reanalyzed into root + future suffix *-se/o-. This created a
distinction between the future form *-se/o- and the aorist subjunctive form *-e/o-. However, a difficulty arose for
intransitive roots. Among root aorists, adding an -s- form was associated with transitivity and was used to create
transitive stems from intransitive ones. Thus, if a future suffix *-se/o- was added to *sta- ‘stand’ it would also signal
a transitive meaning ‘will set up’. To mitigate this problem and maintain intransitive future formation, the
transitivizing effect of the -s- suffix had to be fixed in some way. This fix could be realized by replacing the active
suffix with a middle suffix instead, e.g. στήσεται ‘will stand’. Because of the transitizing effect of the -s- suffix, all
intransitive roots that lacked an active sigmatic aorist required a future middle form instead. This allowed the future
middle form to spread by default to more verbs. See Willi (2018, 449-61) for corroborating data and discussion.
12
Many middle-only verbs are traditionally referred to as ‘deponents’, see §2.2.2 for discussion.

12

form, -(θ)η- is limited to only perfective aspect and is not developed through the rest of the verbal
system (Ellis 2016, 153). The traditional morphological division is in Table 1.
Table 1 Traditional division of voice morphology with λύω 13
Imperfective
Active
Middle/Passive
λύω
λύομαι
λύεις
λύῃ
λύει
λύεται
λύομεν
λυόμεθα
λύετε
λύεσθε
λύουσιν
λύονται
Perfective (aorist)
Active
Middle
Passive
ἔλυσα
ἐλυσάμην
ἐλύθην
ἔλυσας
ἐλύσω
ἐλύθης
ἔλυσεν
ἐλύσατο
ἐλύθη
ἐλύσαμεν
ἐλυσάμεθα
ἐλύθημεν
ἐλύσατε
ἐλύσασθε
ἐλύθητε
ἔλυσαν
ἐλύσαντο
ἐλύθησαν

For the imperfective (including present, imperfect, perfect, and pluperfect paradigms),
one set of forms (-ω) marks active voice, while another set (-μαι) marks non-active, or middlepassive. In traditional terms, these middle-passive forms are capable of expressing either middle
or passive functions. This means that verbs in (1.9) may be used with -μαι morphology to express
either a middle event or a passive event with an external cause.
(1.9)

ἀναιρέω ‘kill, do away with’
ἀπόλλυμι ‘ruin, destroy’

ἀναιρέομαι ‘claim for oneself, adopt (tr.)’
‘be killed (by)’14
ἀπόλλῠμαι ‘perish, die’
‘be destroyed (by)’

According to traditional views, the difficulty arises with respect to the three distinct
morphological forms of the perfective paradigm. There are two non-active forms: sigmatic first

13

Greek voice morphology is fused with subject agreement (person and number). Verbal suffix -ται marks
middle-passive voice and third person singular subject agreement. Alternatively, -νται expresses middle-passive
voice and third person plural subject agreement. See Ellis (2016, 122-59) for verbal morphology charts.
14
The agent may be overtly expressed (The guards were killed by invading forces) or implied without
specifying a particular referent (The ambassador was killed).

13

aorist (-σαμην, -σω, -σατο) and -(θ)η- (-(θ)ην, -(θ)ης, -(θ)η). Traditionally, this tripartite
morphological division is taught as the expression of three distinct voices: One set is active (-σα,
-σας, -σεν). Another set (-σαμην, -σω, -σατο) is middle morphology, expressing middle voice. And
finally, -(θ)η- forms are thought to represent a fully distinct passive voice. Thus, -(θ)η-marked
verbs are expected to appear in passive syntax in which the subject is acted on by someone
else.15 But closer inspection of -(θ)η- usage reveals more diversity than traditional conceptions
suggest. Perfective -(θ)η- in (1.10) behaves much like imperfective -μαι in (1.9).
(1.10) μολύνω ‘defile, cause to be dirty’

ἐμολύνθην

χωρίζω ‘divide, cause to separate’

ἐχωρίσθηv

‘become dirty, defile oneself’
‘be defiled (by)’
‘depart from, separate oneself’
‘be separated (by)’

The challenge of voice phenomena is to understand how a variety of expressions are
connected. To do so, it is fruitful to look beyond syntactic discrepancies to consider the
processes that shape them. This shifts the focus of analysis toward human cognition and the ways
in which we understand the world around us, especially the cognitive and communicative factors
that shape language behavior and linguistic expression (Bybee 2010).
One factor crucial to voice is transitivity. Traditionally, voice is treated as wholly distinct
from transitivity, an assertion made on syntactic grounds. Transitivity is said to relate to the verb
and its object, whereas voice relates to the verb and its subject (Robertson 2006 [1919], 797;
Wallace 1996, 409; Ladewig 2010, 105). But such a narrow definition does little to recognize the

15

With ἀγαπάω in (i) below, active morphology corresponds to active syntax; the subject (my father), as
agent, performs an action on the object (him), as patient. The use of -(θ)η- morphology in (ii) inverts this
configuration, promoting the patient (those who love me) to subject status and supplying the agent (my father) in an
oblique phrase, prototypically ὑπό + genitive case in Greek.
(i) ὁ πατήρ μου ἀγαπήσει αὐτόν
(ii) ὁ δὲ ἀγαπῶν με ἀγαπηθήσεται ὑπὸ τοῦ πατρός μου
my father will love him (John 14:23)
those who love me will be loved by my Father (John 14:21)

14

role of event conception in voice distinctions.16 By way of analogy, consider the role of tense in
grammatical expressions. Tense relates to conceptual distinctions in time. A speaker may
conceive an event as taking place in the past or future relative to speech time. Certain
grammatical expressions provide the means of articulating such distinctions in language. In a
similar fashion, voice is also concerned with event conception. But voice relates to how events
unfold and how participants are involved in the process. If tense indicates distinctions in time
resolving the question of when an event occurs, then voice entails distinctions in process,
regarding how an event transpires (Langacker 1991, 362-71; Shibatani 2006, 221).
This suggests that voice is ultimately concerned with semantic distinctions in event
development. It also ties voice to the notion of semantic transitivity, a conceptual parameter that
pertains to the developmental stages of actions, or how events unfold. In a basic transitive event,
an action is brought about by an initiating force that is transferred to a distinct secondary figure
that then goes through a process of change. This kind of event structure provides a cognitive
frame for understanding voice alternations. Voice distinctions represent departures from the
basic transitive, allowing for different types of oppositions in event structure, pertaining to
various aspects of an unfolding process, and alternate ways of viewing an event. Because actions
have consequences, humans have good reason to make cognitive distinctions in event structure.
Grammatical voice provides form-meaning relationships in language that allow speakers to
16
Event conception refers to a broad cognitive ability allowing humans to effectively interact with our
environment (physical, social). The ways in which we conceive events and organize them into patterns is what
mediates between perceived situations in the world and the way we talk about them in language. Conceptual event
structure, or conceiving events and organizing them into patterns, relies on the notion of cognitive models. These are
integrated conceptual patterns fundamental to human experience. Because they have specifiable properties that
underlie event structure as organized in language, we can investigate human conceptualization through language
typology and analysis, noting how a certain set of meanings are specified together under a single formal expression
in language after language. This reflects some conceptual similarity in how they are understood. Alternatively, a
certain set of meanings may be expressed in different ways, reflecting a notable difference in how they are
conceived (Lakoff 1987; Langacker 1991, 282; Kemmer and Verhagen 1994; Kemmer 2003; Lazard 2003).

15

communicate a variety of meaningful conceptual distinctions in how actions unfold, especially
how participants are involved in the process and how the nature of their involvement alters their
communicative value in the described event (Kemmer 2003, Langacker 2006, Maldonado 2007).
1.2 Motivations in voice
The way we understand events shapes the way we express them in language. Processing
mechanisms like event conception help us to engage in the world by taking vastly different
experiences and understanding them through recurring patterns. For voice, this involves patterns
of interaction between events and participants (Kemmer 2003, 95; Langacker 2006, 115-16).
Historically, these types of mechanisms have received little attention in voice analysis.
Instead, the focus is on syntactic relationships, especially the role of the subject in relation to the
verb. Syntactic constraints are taken as given, without further exploration of how voice
alternations are motivated by cognitive and communicative principles. A semantic account
prioritizes such processing mechanisms by attending to the entire semantic scope of an event:
differences in lexical meaning, participant relations, and semantic and pragmatic functions in
voice. This results in the interaction of three parameters: (a) event development – how events
arise, progress, and come to an end, (b) the way in which participants are related, and (c) the
relative salience of participants in the process (Shibatani 2006, 219), as in (1.11).
(1.11) (a) how events unfold in the flow of energy: how they begin, progress, and end
(b) how participants are related within event development
(c) how their involvement affects the relative salience of participants17

17

‘Relative salience of participants’ focuses internally on a specified event, especially how participants
interact in it and their salience vis-à-vis other participants in the same event. It is fundamentally relational; one
participant may be highlighted relative to a secondary figure or chosen as a singular focus relative to other possible
entities in the event. It is not a statement about a discourse level emphasis/prominence that applies to other states or
events in the larger narrative. Still, the focusing or defocusing of a particular participant in an event can have a
larger discourse function, such as the use of the syntactic passive as topicalizing the patient and defocusing the
agent, especially if the agent is unknown or irrelevant within the narrative (Shibatani 1985, Langacker 2006).

16

The first highlights a contrast in the construal of events, regarding their origin. In the
sentence, My father picked me up and dusted me off, the event’s beginning or starting point, is an
agent external to both the process (the father does not experience the sensation of lift) and the
affected participant (the father and the speaker are distinct participants). Whereas, in I got up and
dusted myself off, the initiation and origin of the event are internal to the person affected. These
differences in English construal are parallel to Greek voice distinctions in (1.12).
(1.12) active ἐγείρω ‘I raise, lift up s.t.’ vs. middle ἐγείρομαι ‘I rise, get up’18
Alternations like (1.12) illustrate a contrast in event origin: external cause vs. internal
energy, shifting the origin of the event from an external, unaffected participant to a participant
internal to the process of change itself.
Relationships among participants also underlie how event construal affects language
structure. Active διαμέριζω ‘distribute’ in (1.13) profiles a process where an agent (people of the
church) transfers a theme (their possessions) (accusative object) to a recipient (those in need)
(dative indirect object). Each participant plays a distinct role in event development. The event’s
energy moves with the transferred possessions from the agent to a distinct recipient.
(1.13) διεμέριζον αὐτὰ πᾶσιν καθότι ἄν τις χρείαν εἶχεν
they would distribute them [their possessions] to all, as any had need (Acts 2:45)
(1.14) διαμερίζονται τὰ ἱμάτια αὐτοῦ
they divvied up his clothes (Mark 15:24)
But the middle form of the same verb in (1.14) offers a different construal in participant
relations, mapping the transference frame with three participants onto a two-participant event.
This has the effect of profiling an event that centers on a single set of primary participants who
are construed as both agent and recipient. The event begins with the agent (soldiers) who acts on

18

‘He got up from the table and removed his robe’ ἐγείρεται ἐκ τοῦ δείπνου καὶ τίθησιν τὰ ἱμάτια (John 13:4).

17

a patient/theme (clothes) (accusative object), but rather than terminating in a distinct recipient
who receives the clothes, the soldiers are the recipients. Two roles are conflated onto one
participant (Kemmer 1993, 78-81). The soldiers, as a group, play the starting point (agent) and
endpoint (recipient) of the event; the transferred object stays with the soldiers. Participants in
(1.14) share a different relationship in event structure as compared to participants in (1.13).
The relative salience of participants emerges as a natural by-product of how participant
relations are construed. The middle in (1.14) profiles an event that develops only around the
primary figure without extending to a distinct recipient. This alters the scope of attention as
compared with (1.13). Rather than extending attention from an initiating figure to a distinct
recipient, the locus of attention remains on the primary figure as the final recipient of the action.
As a result, each construal highlights different facets of a profiled relationship by changing the
direction of the action. The active profiles an event that extends outward from its initiating point,
highlighting the agents’ activity as givers who distribute their possessions to those in need. The
middle profiles an event that develops inward. The action focuses on those who initiated it. This
has the effect of highlighting the recipient role of the soldiers. For both active and middle, the
agent is the primary figure highlighted as the doer of the action, but only the middle draws
attention to the salience of the soldiers as starting point and endpoint of their own action.
These factors – event origin/process, relationships between participants, and the relative
salience of the participants – all implicate meaning-oriented distinctions rather than syntactic
ones that traditional analyses tend to focus on, independent of rationale. By connecting voice to
semantic relations among events, we more readily tie voice to notions of semantic transitivity,
that is, patterns of action and interaction among participants, rather than syntactic rules. Viewed
this way, transitivity becomes a continuum or gradient scale for various semantic event types;

18

each of which can be more or less transitive, depending on proximity to the prototypical
transitive event (Langacker 1991, 282-329; Næss 2007, 12-15; LaPolla et al. 2011) and voice
relations are grounded within this continuum.
The transitive prototype represents a highly transitive event where a volitional agent
purposely acts on a distinct patient, causing a physical change of state/location in the patient, as
in (1.15) with active ῥίπτω ‘throw’. One participant (a group of sailors) supplies the initiating
force; the other (the ship’s gear) absorbs the energy and is changed by it (Kemmer 1993, 50-51).
(1.15) τῇ τρίτῃ αὐτόχειρες τὴν σκευὴν τοῦ πλοίου ἔρριψαν
on the third day with their own hands they threw the ship’s gear overboard (Acts 27:19)
The active encodes an event that is closer to the transitive prototype and thus has a higher
degree of transitivity. The primary figure causes a change to occur in a secondary participant.
The middle of the same verb depicts an event that departs from the prototype. In (1.16), the
primary figure experiences being tossed about, rather than causing it to happen to another.
(1.16) ὁ γὰρ διακρινόμενος ἔοικεν κλύδωνι θαλάσσης ἀνεμιζομένῳ καὶ ῥιπιζομένῳ
the one who doubts is like a wave of the sea, driven and tossed about (James 1:6)
Voice alternations divide up the space of transitivity, categorizing semantic event types in
their relation to or departure from the transitive prototype. Events deviate from the prototypical
transitive in various ways, as in (1.12) and (1.14). These departures are coded in language by
grammatical means with an overt voice pattern, such as middle morphology, in order to signal
their status as distinct from the default active (Maldonado 2007, 864).
Voice systems form a set of relations in language, linking semantic event types along a
scale of transitivity with different conventionalized expressions. Active voice encodes the
transitive prototype, or high semantic transitivity, with the basic nominative-accusative pattern in
a language like Greek (Shibatani 2006, 220; Langacker 2006, 123). In this basic alignment, the
event origin and endpoint are filled by two fully distinct participants. The middle provides an

19

alternate construal. Events that depart from the transitive prototype with respect to the conceptual
parameters above receive middle marking in order to formally signal a semantic shift in how the
event unfolds. Among middle event types, origin and endpoint roles are filled by the same
participant. Situating voice phenomena within human behavior, especially with regard to event
development and semantic transitivity, not only provides for cognitive motivations in voice, but
also its role as a relational tool in communicative contexts.
One common assumption in the study of language is that the meaning of a linguistic
form, e.g. middle morphology, is given, and that the context in which it is placed is the variable
element that further clarifies or elucidates its grammatical value. But due to the nature of human
interaction, the process should really be considered from the other way around. In real-world
communication, speakers do not have a choice regarding the context in which they speak or write
(Bache 1997, 103). But they do have a choice in the form of grammatical expression they use to
talk about it. In real-world communication, the context is given, and the form of the linguistic
utterance is the variable element used to interpret it (LaPolla 2003, 120).
If language develops in communicative activity, then grammatical categories, such as
voice, represent usage patterns that are conventionalized over time as part of the grammatical
makeup of the language. Their function in communication rises through a process of repetition
and recognition: As they are repeatedly used for limiting the interpretation of events in particular
ways, this interpretation becomes a recognized part of language convention and shapes continued
use (Harris 1998, 27-45; Langacker 2006, 109). Thus, our study of the middle voice concerns
how it functions within the voice domain to constrain the interpretation of events.
In summary, voice is integrated in the construal process – the human ability to consider
events in differing ways and express them in language accordingly (Langacker 1987, 294; 2006,

20

118-121). Within the grammatical resources of a language like Greek, an event can be coded in
multiple ways with alternate grammatical devices that reflect how a speaker conceives it as an
unfolding process. Voice oppositions embody different ways of viewing how events unfold. A
speaker chooses a way to view an event, adopts it for the purpose of communication, and
suggests it to their audience by means of a shared grammatical structure (Verhagen 2007, 48-81).
The three parameters in (1.11) may be subsumed under two principle motivations for the
larger construal process. The first, energy flow, captures the first two parameters, namely how
events unfold and how participants are related within the process. The second, focus of attention,
captures the third parameter: how participant involvement affects the salience of different facets
of event construal (Givón 1984; Langacker 2006, 118-19). Departures from the prototypical
transitive event regarding the interplay of these motivations give rise to voice alternations:
• Energy flow: This parameter has to do with changes in semantic event types with respect
to how energy is transferred in an event, particularly pertaining to adjustments in energy
source, progress, and endpoint (Shibatani 2006, 219).
• Focus of attention: This involves a visual metaphor in how attention can shift to different
parts of an event. Adjustments in attention motivate adjustments in voice. In this way,
voice reflects attentional focus (Langacker 1987, 115-17; Croft and Cruse 2004, 40).
o A natural consequence of shifting the center of attention in an event is a
change in the relative salience of event participants (Langacker 2006, 121-27).
Broadening the scope of voice analysis to include shifts in the construal process also
subsumes the popular notion of subject affectedness (Allan 2003, Campbell 2015). But it does so
with an emphasis on affectedness as a relational principle, involving interdependent elements.
Subject affectedness should not be treated as a static primitive as present or not. Rather, it is a
matter of degree, as transitivity is, based on how events unfold. Affectedness concerns: (1)
change along a scale, a participant can be more or less affected by their involvement, and (2) the
nature of its endpoint. The more specific a verb is about a participant’s progress along a scale of

21

change, the higher the degree of affectedness for that participant.19 Verbs that involve a change
of state entail a higher degree of affectedness than verbs that are unspecified for change (Beavers
2011, 356-62). This is a difference between high semantically transitive verbs (explode, destroy,
kill) which entail a specific result state in contrast to verbs that are lower on the transitivity scale
(follow, ponder, see) which do not entail any necessary change for the endpoint figure. When
affectedness for the subject is considered in relation to the middle, these parameters come in to
play. Yet since it is a gradable and relational property that involves the nature of the predicate
itself, subject affectedness may or may not be readily recognizable in any given middle event.
Viewing the middle as a multifunctional category grounded in human cognition allows us
to engage the construal process and the nature of event categorization. This approach seeks to (1)
draw attention to shared cognitive processes and interpretive behavior that shapes the character
of the middle, and (2) capture relations among subtypes and instances of use in a motivated way
without positing separate rules that operate on underlying language structure.20 Exploring
external motivations for language recognizes the integrated nature of the system; the same
principles that shape difference also shape unity (Harris and Wolf 1998, Verhagen 2002).
1.3 Organization
The remainder of the chapters is organized as follows: Chapter 2 places the Greek middle into a
cross-linguistic context, noting similarities and differences with respect to the behavior of voice

19
This highlights the importance of lexical semantics in voice. The specific semantic nature of the
predicate has a direct impact on the grammatical devices used with it.
20
See Lavidas and Papangeli (2007, 117-21) for a review of various approaches to the Greek middle
(semantic, syntactic, morphological). Lavidas and Papangeli approach voice syntactically, suggesting that deponent
middle verbs ought not to appear in clauses with an accusative object. Where this happens, they posit a
morphological feature specification that operates on underlying language structure relative to a certain set of verbal
stems. This is done in order to account for middle-marked verbs that do not meet certain syntactic qualifications for
the middle. These morphological features apply to verbal stems outside of syntactic structure and are realized in a
random way, possibly relying on idiosyncratic verbal properties (120).

22

alternations across languages and identifying where Greek middle morphology fits in among
them. Chapter 3 considers the Greek middle in diachronic perspective, especially with regard to
two particular processes of change. The first pertains to lexicalization in voice and the presence
of formal and semantic idiosyncrasies. The second relates to a process of grammaticalization, as
illustrated through the rise of perfective -(θ)η- morphology in Greek, tracing its origins and
integration into the Greek voice system. Chapter 4 advances a semantic analysis of the various
middle event types, introducing notions of conceptualization and construal, with the scale of
transitivity as a conceptual grounding for distinctions in voice. Each semantic event type is
discussed in turn, along with relations and distinctions among them. Chapter 5 provides final
remarks on the semantic event types that constitute the middle domain in Greek.

23

2. Middle typology
Chapter 2 examines the behavior of the Greek middle in the context of cross-linguistic patterns
in voice. To begin, I explore the fundamental differences between derived voice systems and
basic middle systems (§2.1). Traditional conceptions of voice describe the Greek middle
according to expectations that follow from the classical active-passive constrast, which not only
misrepresents the grammatical behavior of Greek middle voice alternations, but also obscures the
functions middle morphology serves within the Greek voice system.
In order to identify the semantic behavior that occurs within middle voice, we need to
place the description of the Greek middle into its typological context and clarify the ways in
which the Greek voice system behaves in kind with other languages that express middle voice.
Identifying these semantic alternations in Greek voice provides a more accurate description of
the kinds of patterns that drive the expression of middle morphology.
To that end, section 2.2 highlights semantic regularities in the Greek middle that are often
overlooked by traditional syntactic accounts, with particular attention to the semantic restrictions
that exist among anticausative alternations in voice as well as the distributional patterns of
middle morphology across recurring semantic event types. Such patterns in Greek are fully
consistent with cross-linguistic norms among middle voice systems. Finally, section 2.3 offers a
cross-linguistic survey of common middle types, illustrating the remarkable functional similarity
that exists across middle systems along with the ways in which Greek voice fits in among them.
Typological research aids in the process of language description by way of comparison
and contrast. Many languages express voice contrasts that are not based on the active-passive
opposition, do not resemble it, and are in fact typologically distinct from it (Kemmer 2003, 89118; Tsunoda and Kageyama 2006).

24

Cross-linguistic evidence points to the diversity of voice phenomena, highlighting
patterns across the world’s languages. In the wider context of language typology, we can locate
Greek with respect to other languages that show similar, or markedly different, voice patterns
(Quirk et al. 1985, 159-71; Klaiman 1991, 44-109; Fox and Hopper 1994). The following section
focuses on two typologically distinct voice types in order to demonstrate key differences: the
derived active-passive system in a language like English and the basic middle system in Greek.
2.1 Voice systems: derived vs. basic
English exhibits an active-passive alternation. Greek, on the other hand, represents a distinct
voice type with an active-middle(passive) alternation.21 One key difference stands out between
them. The former is a syntactically derived system, i.e. traced from a source, and the latter is a
basic system, i.e. not traced from a source (Klaiman 1991, 23-24).
Greek fits with many other languages, e.g. Turkish, Hungarian, and Tamil, in expressing
middle voice (Klaiman 1991; Kemmer 1993). Other voice systems include derived passive
systems in English and Lango (E. Sudanic, Uganda), anti-passive systems in Chukchi (N.
Chukotko-Kamchatkan, Russia) and West Greenlandic (Eskimo, Greenland), inverse systems in
Native American languages like Navajo (Athapaskan), and Philippine systems in Tagalog,
Cebuano, and Ilokano (Foley and Van Valin 1984, 135; Kozinsky et al. 1988, 652; Van Valin
and LaPolla 1997, 295-8; Farrell 2005, 91; Keenan and Dryer 2007, 358-59).
Ultimately, research in this area makes two simple claims: (1) different languages express
different distinctions, and (2) we cannot rely on one type of system to define all others. Doing so
21

Recall that the passive alternation in Greek is expressed by the same set of voice morphology as the
middle. Greek does not distinguish a distinct morphological passive but includes both middle and passive functions
within a single marking pattern. Though -(θ)η- morphology is traditionally distinguished as a separate marker of
passive voice, it should actually be included with the rest of middle-passive morphology, able to express middle and
passive functions with a single formal expression. Section 3.2 considers this in detail.

25

diverts attention and distorts expectations for how a system ought to behave. This is because
languages reflect a variety of types that evince different categories from those in the classical
active-passive opposition (Shibatani 1988, 2004; Dixon and Aikhenvald 2000; Farrell 2005).22
One defining feature of a syntactically derived system is that the marked passive voice
entails a syntactic remapping from the basic active structure to a non-basic, or derived,
configuration. The passivization process in (2.1) remaps the participants in the clause so that the
object of the active is reconfigured as the subject of the passive.
(2.1)

(a) Joshua consumed all the champagne
(b) All the champagne was consumed by Joshua
For the active in (a), the subject has the semantic role of agent (doer) and the object has

the role of patient (receiver/undergoer). In the passive in (b), the opposite is true. The patient is
given subject status and the agent is demoted to a non-argument oblique phrase (by Joshua). The
passivization process derives a marked voice structure from the basic active by taking the patient
object of the active (all the champagne) and remapping it to the subject relation in the passive.
Passivization involves syntactic derivation: a reassignment of semantic roles (agent, patient) and
grammatical relations (subject, object). The derived passive in (b) has a patient subject
(champagne); the default active in (a) has an agent subject (Joshua).
Several grammatical consequences follow from this type of alternation. The first is that
passivization (i.e. deriving a syntactically marked voice from a basic structure) requires basic
transitive clauses. All marked voice structures (i.e. passives) must have an active counterpart.

22

Intuitions among Greek scholars that voice categories in English do not fit those in Greek reflect this fact
(Gildersleeve 1900, 64n145; Lyons 1968, 373; Wackernagel 2009, 160; Porter, Reed, Brook O’Donnell 2010, 125).

26

There cannot be passive-only verbs in this type of system. The passive category can only arise by
syntactic derivation from a basic structure; all passives have an active as their source.23
Second, because voice alternations in derived systems are based on the transitivity of a
clause, only syntactically transitive clauses can be passivized. Intransitives, e.g. Jeanette danced,
with only one argument are prohibited from voice alternations; there is no direct object relation
to promote to subject status. The passive alternation hinges on syntactic transitivity, specifically
the presence of two (syntactically required) arguments. Intransitive (one-argument) verbs, by
their very nature, are barred from voice alternations because there is no second argument from
which to project the event. This means that for intransitives, whether they involve a volitional,
agent-like subject (Jeanette danced) or a non-volitional, patient-like subject (Jeanette fell), the
semantic difference does not trigger a difference in formal expression. All intransitive verbs are
expressed in the same default active pattern; they cannot be passivized (Klaiman 1991, 4-22).
Finally, voice alternations in derived systems are expected to be semantically neutral.
Voice remains the property of the formal coding of a clause and is not identified as a semantic
phenomenon. The semantic content of the action is regarded as unaffected; only the alignment of
the subject and object to the verb changes, with no semantic contrast in view. The semantic
relationship between active and passive remains the same despite a change in voice. Both (2.1)
23

While passivization is a fairly regular process in English (most transitive verbs can be passivized), there
are still semantic limitations on its grammatical expression. Some verbs, whether in general or with a particular
sense, do not alternate between voices; their use is restricted to active only: (a) This resort boasts high standards
(*High standards are boasted by this resort); (b) These shoes fit my feet (*My feet are fit by these shoes); (c) Her
suitcase weighs fifty pounds (*Fifty pounds is weighed by her suitcase). Passive restrictions like these are based on
the semantics of the verb, tied to their low degree of transitivity, especially with regard to affectedness of the object
participant. Objects that are less affected cannot be passivized in English. This is a difference between a highly
affected object in Ed crushed the ice (The ice was crushed by Ed) vs. a virtually unaffected object in Micah has thick
hair (*Thick hair was had by Micah). The same principle can be illustrated with more ambiguous examples, as in
(active) Oliver climbed the stairs vs. (passive) The stairs were climbed by Oliver. Even though the subject (Oliver)
volitionally performs an action on the object (stairs), the object remains relatively unaffected by the action of
climbing. While this passive alternation may be grammatical, it is still semantically odd and less likely to appear in
communicative contexts (Ward, Birner, and Huddleston 2002, 1432; Shibatani 2006, 219; Croft 2012, 253).

27

and (2.1) have the same propositional content. The meaning of the verb remains the same, with
modification only in the syntactic alignment of arguments in the clause (i.e. the relationship of
the subject to the verb) (Shibatani 2006, 236).
To this derived system, we can contrast the behavior of a basic middle system (Klaiman
1991, 44-109). Cross-linguistically, middle systems have a primary division between two formal
types: a basic unmarked active and a marked counterpart. Yet its grammatical behavior differs
from that of the derived system. First, middles are not syntactically derived like the passive.
Middle voice entails no reconfiguration of arguments from a basic to a non-basic alignment. The
choice between active and middle does not rely on relocating clausal participants with respect to
their grammatical roles. The active-middle alternation between αἱρέω ‘take s.t.’ and αἱρέομαι
‘choose s.t.’ provides no reason to consider one as basic and one as derived; no syntactic
derivational process acts on the active to produce the middle (Klaiman 1991, 24). In both, the
subject performs the action. Semantic roles (agent, patient) and grammatical relations (subject,
object) remain the same. The primary difference lies in the semantic shift of the verb itself. In the
active ἅπτω ‘ignite s.t.’ vs. middle ἅψωνται ‘touch s.o./s.t.’, there is no reconfiguration of
participants that derives the marked from the unmarked voice. These and other examples
illustrate that middle systems are primarily based on the semantic organization of events, not on
the syntactic reassignment of clausal arguments (Klaiman 1991, 44; Shibatani 2006, 236).
This typological principle impacts the kinds of relations we ought to expect between
middle forms and the existence of active counterparts. Greek -(θ)η- morphology plays more than
just a syntactic passive role in the Greek voice system, and in fact its functions overlap with
other middle-passive (-μαι) morphology in Greek (see §3.2). Instead of expecting all -(θ)η- verbs
to have an active (source) counterpart as would be the case in the English derived system, some

28

verbs will inflect for one voice and not the other because they are not created by a derivational
process from the active. A given verb with -(θ)η- morphology does not presuppose an active
source as its basis. This applies equally to -μαι morphology as well. The widespread presence of
middle-only verbs attests to this fact (see §2.2.2) (Kemmer 1993, 22-23; Klaiman 1991, 97-104).
Verbs without a contrasting active counterpart are merely by-products of a system that does not
syntactically derive one voice from another. Both active and middle voices are basic with no
need for derivational reassignment of the roles in the clause. Both passiva tantum (verbs with the
-(θ)η- form and no active) and media tantum (middle-only verbs with no active), are basic, i.e.
not derived via a syntactic process from the active, and must be understood apart from
passivization (Wallace 1996, 441; Allan 2003, 2-3; Maldonado 2009, 69-111).
Related is the issue of transitivity in middle systems. Because the active-middle contrast
is not directly correlated with syntactic transitivity, middle verbs are not restricted to a particular
transitivity structure. Middles appear as both one- (intransitive) and two- (transitive) argument
verbs. In contrast to the derived passive alternation, voice in Greek is not directly correlated with
reducing the number of syntactic arguments in a clause.
Recall that in the derived active-passive system, voice is based on reducing the number of
arguments from two in the basic active voice (e.g. Dave bounced the ball) to one in the marked
passive voice (e.g. The ball was bounced [by Dave]). As a result, there are no two-argument
passives possible in a derived active-passive system. But in middle systems, marked voice
constructions appear in either intransitive or transitive predications with the direct object left
intact (Klaiman 1991, 106).24 When -μαι or -(θ)η- forms appear in transitive clauses, this does not
break a syntactic valency rule, but merely means that both alternations involve actions in which
24

The semantic relationship between transitivity and voice is discussed in §4.1.

29

energy is transferred from one participant to another. The semantic nature of the event impacts
how it is expressed in syntactic structure.
For example, the causative/anticausative contrast in Greek actually does correlate with a
change in valency: λυπέω ‘make sad’ vs. λυπέομαι ‘be/become sad’. The alternation in voice
hinges on a difference in how the event is brought about. In the active, a distinct external agent
brings about a change of state in a second participant, hence two arguments in the active. The
anticausative middle, however, involves just one participant who experiences a change of state
without an explicit external cause. Because of this change in how the event is brought about,
there is no semantic motivation for a two-argument structure, thus the middle is expressed as a
one-argument clause. In contrast, an alternation like ἀναιρέω ‘do away with’ vs. ἀναιρέομαι ‘take
up, claim’ retains a second argument with no change in valency. Both actions involve two
participants; an agent does something to cause a change of state/location in a patient. The
difference is in the development and endpoint of the event. In the active, the action develops
away from the primary figure and ends with a distinct object. In the middle, the arc of the action
develops toward the primary figure and terminates with the one who started it. Here, a change in
voice does not correlate with a change in the number of arguments, i.e. valency reduction.
Finally, middle alternations are not expected to be semantically neutral. In the activepassive contrast, voice is defined as a syntactic coding mechanism that alters the alignment of
clausal arguments. No semantic shift is in view between Dave bounced the ball vs. The ball was
bounced by Dave. Only the arguments in the clause are reassigned; the semantic action remains
the same. Regardless of grammatical status in the clause (subject, object), the semantic role of
Dave remains the agent, the ball remains the patient, and the developmental action of ‘someone
bounces an object’ remains intact. What is not expected in this type of derived passive is a

30

semantic shift in how the event takes place, such as an anticausative contrast: Dave bounced the
ball vs. The ball bounced. In this case, the event development actually changes from ‘someone
bounces an object’ to ‘an object bounces’. But in a middle system, oppositions like the semantic
shift in the anticausative alternation are common. This is why Greek grammars frequently note
that sometimes the contrast between active and middle changes the lexical meaning of the verb
(Wallace 1996, 416; Robertson 2006 [1919], 804). This behavior may seem odd from the
perspective of a syntactic view of voice, but in a middle system it is an anticipated attribute.
Placing two voice systems side by side allows us to note the differences in grammatical
expression and expected behavior. In a syntactically derived passive system, voice alternations
entail a structural remapping of participants in the clause. Voice is predicated on a syntactic
derivational process from a basic structure to a derived one. Because of this, (1) there are no
passive (or middle)-only verbs. And (2) voice is a valency reducing operation; intransitive (oneargument) verbs cannot be passivized. Finally (3), a change in voice only alters the grammatical
alignment of participants in the clause; the meaning of the verb is left unaffected.
In contrast, a number of common attributes among middle systems are manifest in Greek:
(1) voice alternations are basic, non-derived structures, and the presence of non-alternating verbs
is perfectly natural to the system, (2) middle alternations are predicated on shifts in semantic
event types not on syntactic transitivity; thus, they do not necessarily result in a change in clausal
valency, i.e. argument reduction. And (3) alternations are not expected to be semantically
neutral; they entail a semantic shift in the meaning of the verb, i.e. the nature of the event type.
While these attributes in Greek are difficult to explain from a syntactically derived point of view,
they are easily explainable from the perspective of a basic middle system that relies not on
syntactic reassignment rules but on semantic shifts in event types.

31

Thus, the descriptive problems in the Greek middle are due more to a misguided use of a
derived passive system than to Greek voice operating differently than typologically expected in a
basic middle system. Greek, as a middle system, behaves in kind with other typologically middle
systems. It cannot be adequately understood without reference to changes in semantic event
structure.25 Section 2.2 picks up this theme, illustrating semantic patterns in Greek middle
morphology that are often neglected among traditional (morpho-)syntactic views of voice.
2.2 Syntactic accounts and semantic regularities
Contemporary methods of analysis reflect the historical foundations on which they are built. One
of the most prominent practices in voice comes from the study of classical literary languages,
especially Latin, but also Greek and Sanskrit (Indo-Aryan), and their application to modern
European tongues, e.g. French, German, and English (Klaiman 1991, 2; Richards and Rodgers
2014, 4-7). Modern accounts of voice that stem from the classical tradition tend to treat the
active-passive dichotomy as the main contrast, with the middle serving as an addendum to an
already established framework (Crystal 2003, 495; Shibatani 2004). Many traditional Greek
grammars assume this classical foundation, treating the middle in relation to syntactic
expectations that follow from active and passive structures.
Among traditional conceptions, voice is identified as a formal category of the verb or
clause, with a particular interest in how voice alters the subject argument and its semantic role
status in relation to the verb (i.e. whether it is an agent or patient of the action) (Wallace 1996,
408). For classical languages like Greek and Latin, a change in verbal morphology is expected to
correlate with a change in the structural alignment between semantic roles (agent, patient) and
25

As a descriptive approach, this applies to both the Greek middle as well as to the passive. Greek behaves
in kind with a number of other middle systems that also include the passive function in the middle-marking domain
(see §2.3). Recall that imperfective -μαι morphology expresses middle and passive functions, as noted in §1.1.2.

32

grammatical relations (subject, object). This creates a form-to-function mapping between
morphology and syntax, with each morphological form defined by its role in syntactic structure
(Klaiman 1991, 2-31; Shibatani 2004, 1146).
And yet neat correspondences between a morphological form and a distinct semantic role
for the clausal subject (i.e. agent [active], patient [passive], beneficiary [middle]) make up only a
subset of voice phenomena in Greek. Most middles are not reflexive and since there is no distinct
role to afford them, traditional approaches ascribe a ‘self-interest nuance’ to the middle (Mounce
2009, 228). The subject is said to be intimately involved, directly influenced, or especially
emphasized (Dana and Mantey 1927, 157; Smyth 1956, 390; Young 1994, 134; Wallace 1996,
414-15; Lavidas and Papangeli 2007, 100; Ladewig 2010).
Such accounts may have an appeal, since they seek to offer a separate syntactic function
for each morphological form, but they sustain a rather skewed picture of voice in Greek. This is
due to two contributing factors: (1) the practice of looking at voice morphology through a
syntactic rather than semantic lens; (2) the tendency to treat Greek analysis outside of any crosslinguistic context. Or, to put it another way, the exclusive focus on voice as a morphosyntactic
pairing has led to a relative neglect for deeper semantic regularities that play out in voice, both in
Greek and across languages with middle systems (Kemmer 1993, 23-24; Shibatani 2006). To
address such semantic regularities, two typological patterns are worth considering. Section 2.2.1
concerns the multifunctional nature of voice morphology. Section 2.2.2 follows with a discussion
of the relationship between voice and lexical semantics.
2.2.1

Anticausatives and passives

One recurring pattern across languages is that elements classified as passive markers (i.e. they
are used to express passive syntax) more often than not express a number of other functions as

33

well, including anticausatives, impersonals, potential passives, fientives, resultatives, etc.
(Shibatani 1985, 825-30; Haspelmath 1990, 37-49; Givón 2002, 207-14). This fact alone begins
to challenge traditional accounts that treat -(θ)η- morphology as uniquely passive in function. But
in particular, the special treatment of the passive voice as a distinct realization in the argument
structure of a clause disregards the grammatical continuum that exists among voice
alternations.26 Consider the relationship between passives and anticausatives in (2.2).27 Both
serve as alternate construals of the same causal event (Kulikov 1998, 140; Croft 2012, 253).
(2.2)

(a) Cody stopped the music (b) The music stopped
(c) The music was stopped (by Cody)
The relationship between (a) and (b) is not a simple valency reduction. This would be a

difference between ‘Cody stopped the music’ vs. ‘Cody stopped’, in which the transitive object
(patient) is removed to produce an intransitive structure that shares the same subject argument
(Cody).28 Instead, the anticausative in (b) highlights a very specific kind of detransitivizing
process, one in which the undergoer, or patient, of the transitive (‘stopped [x]’) is given subject
status in the alternate structure (‘[x] stopped’). In fact, the same operation is true in (c); the
object of the transitive becomes the subject of the passive. For both (b) and (c), the subject
(music) undergoes a change of state; i.e. it ceases to play. The key difference between the two
lies in the presence or absence of an external force in bringing about the change. In a passive
construction, the agent remains conceptually present, though its role is downplayed, either
26
The passive is treated as unique among voice types in regard to its argument realization. According to
traditional views, the passive (as opposed to active or middle) is the only voice type that requires a realignment of
the arguments in the clause so that the agent is not realized in its default role as subject. Note that middle voice, as
defined in teaching grammars, keeps the agentive participant in the default subject position so that there is no
argument reassignment necessary. The passive is treated as unique in its expression of the subject as patient.
27
The term ‘anticausative’ is used in this section in a purely functional fashion, meant to draw out crosslinguistic semantic patterns and is not meant as a commentary on any specific formal derivational processes.
28
In the alternation ‘Cody stopped the music’ vs. ‘Cody stopped’ the object is removed. Since the process
entails no realignment of clausal arguments, the agent (Cody) remains intact as grammatical subject in both.

34

expressed as an oblique phrase or left implied without overt coding. Thus, (c) remains an
essentially causal event – an agent causes a patient to do something. The anticausative in (b),
however, removes the agent entirely. As a non-causal construal, the event takes place
spontaneously without direct intervention from an outside force (Haspelmath 1987, 7; 1993, 90).
A crucial challenge for traditional accounts is that a number of languages express both
alternations with a single form. In Amharic, the verbal prefix t(ə)- applies to two-participant
events: səbbərə ‘break (tr.)’, kəffətə ‘open (tr.)’. The resulting expressions (tə-səbbərə and tə-

kəffətə) may be either anticausative (‘it broke’, ‘it opened’) or passive (‘be broken (by)’, ‘be
opened (by)’) (Amberber 2000, 315). In Greek, the same pattern holds for perfective -(θ)η- in
(2.3) as well as imperfective -μαι in (2.4). Thus, ἀνεῴχθη, in contrast to ἀνοίγω, may be
construed as anticausative ‘it opened’ like (2.2), or passive, ‘it was opened’ like (2.2).29
(2.3)

ἀνοίγω ‘open [x]’

ἀνεῴχθη

σβέννυμι ‘extinguish [x], put out’ ἐσβέσθη
(2.4)

ἀπόλλυμι ‘ruin, destroy [x]’
ῥήγνυμι ‘break, burst [x]’

‘[x] opened’
‘be opened (by)’
‘[x] went out, died out’
‘be extinguished (by)’

ἀπόλλῠμαι ‘[x] perishes, dies’
‘be destroyed (by)’
ῥήγνῠμαι ‘[x] breaks, bursts’
‘be broken (by)’

Another challenge for traditional approaches is the semantic nature of such alternations.
They do not apply systematically across all verbs (as purely morphosyntactic phenomena) but
are subject to semantic restrictions – a fact that receives little attention in syntactic accounts. The
lexical semantics of a given verb (either in its general use or in a specific sense) constrain the
interpretation of an event such that one type of alternation is more likely than another. Between
29

See Alexiadou (2010, 185) for Modern Greek examples of the same passive/anticausative pattern.

35

the two, the passive remains an essentially two-participant construal, whether or not the external
force is overtly coded, whereas the anticausative is a one-participant event that occurs through
internal means rather than external causation (Levin and Rappaport Hovav 1995, 90-91). For this
reason, not every verb can form an anticausative; certain restrictions apply to its formation.
The first specifies that the action must involve a change of state (or at least some degree
of affect) for the undergoer/patient of the action. This excludes events like watch, thank, or touch
because they do not depict actions that transfer force to a patient, resulting in a change for that
patient. Just because a verb involves two participants, it does not follow that it can form an
anticausative alternation. A verb like ἀναγινώσκω ‘read’ does not depict a change of state for the
patient and thus forms a passive rather than anticausative construal: ἀναγνωσθῇ ‘[x] was read’.
The second restriction specifies that verbs that allow anticausative alternations must be
able to conceivably take place without direct initiation from an external cause. Attempting to
conceive of events in (2.5) as alternating in the same way as in (2.3) is a fairly awkward process.
Verbs in (2.5) are more agent-oriented in meaning. They require an external source of energy in
order to come about and cannot take place as automatic, spontaneous, or internal changes. The
same is true for verbs that imply the use of a specific tool or mechanism (e.g. cutting, biting,
building). Events in (2.5) and (2.6) allow only a passive alternation since no anticausative
construal is likely available (Haspelmath 1987, 15; 1993, 93; 2016, 36).
(2.5)

σταυρόω ‘crucify’
ἐσταυρώθη ‘be crucified (by)’ (*[x] crucified)
ἀποφέρω ‘take away’
ἀπηνέχθη ‘be taken away (by)’
ἁγιάζω ‘sanctify, make holy’ ἡγιάσθη ‘be sanctified (by)’ 30

30

Imperfective -μαι morphology produces the same passive alternation: σταυροῦνται ‘be crucified (by)’,
ἀποφέρονται ‘be taken away (by)’, ἁγιάζεται ‘be sanctified (by)’.

36

(2.6)

ἐκκόπτω ‘cut off’
ἐξεκόπη ‘be cut off (by)’ (*[x] cut off)
οἰκοδομέω ‘construct, build’ οἰκοδομήθη ‘be built (by)’
λατομέω ‘hew out of rock’ ἐλατομήθη ‘be hewn, be trimmed (by)’
Alternatively, the anticausative construal is the more likely pattern for changes that can

occur on their own, or in fact often do occur as spontaneous or internally induced (rather than
externally caused) events, as with perfective -(θ)η- in (2.7) and imperfective -μαι in (2.8)
(Kemmer 1993, 21; Levin and Rappaport Hovav 1995, 90–98; Shibatani 2006, 229-30).
(2.7)

Causative
ἐξυπνίζω ‘arouse [x]’
κοιμάω ‘put [x] to sleep’
ἵστημι ‘place, set, stand [x]’
κατακλίνω ‘lay [x] down’

Anticausative
ἐξυπνίσθη ‘[x] woke up’
ἐκοιμήθη ‘[x] fell asleep’
ἐστάθη ‘[x] stood, stood up’
κατεκλίθη ‘[x] sat down, reclined’

(2.8)

ἐγείρω ‘raise [x], lift up’
(καθ)ίζω ‘seat [x]’
εὐφραίνω ‘make [x] glad’
λυπέω ‘vex [x], distress’

ἐγείρομαι (ἠγέρθην) ‘[x] rises, gets up’31
(καθ)ίζομαι ‘[x] sits, takes a seat’
εὐφραίνομαι (εὐφράνθην) ‘[x] rejoices, celebrates’
λυπεῖται (ἐλυπήθην) ‘[x] grieves’

This suggests that the formation of passives and anticausatives in Greek is largely based
on differences in semantic event types. Certain verbs, in virtue of their lexical meaning, are
unlikely to form anticausatives, but will only be interpreted as passives with an implied external
initiation present in the conception. It should be noted, however, that the anticausative formation
still applies to a range of semantic types. Note the difference between a more patient-like change
of state (ῥήγνῠμαι ‘break’) in (2.4) vs. a more agent-like one (καθίζομαι ‘sit’) in (2.8). In each, the
contrast is between events that are externally caused (‘break s.t.’, ‘seat s.o.’) vs. those that are not
(‘s.t. breaks’, ‘s.o. sits’). In the causal construal, an agent, as the head of the action chain, causes
a second figure to change. In contrast, the anticausative construal for ῥήγνῠμαι and καθίζομαι
originates with the one going through the process of change (‘[x] breaks’, ‘[x] sits’). Where they

31

The anticausative construal does not necessarily preclude a passive alternation. Verbs like ἐγείρομαι may
still be used to indicate a passive construal in which someone or something was raised up by an external force.

37

differ is in volitionality. If an event involves an animate participant who typically performs an
action by his/her own internal energy (stand, sit, walk), then the subject of the anticausative acts
volitionally as an agentive participant (καθίζομαι ‘sit’). If an event involves a change of state that
is typically non-volitional, then the anticausative subject is more patient-like. This occurs with
animate (ἀπόλλῠμαι ‘perish, die’; ἐξυπνίσθη ‘wake up’) and inanimate entities (ῥήγνῠμαι ‘burst’,
ἐσβέσθη ‘die out’ [of flame]) (Haspelmath 1987; Shibatani 2006, 230).
Yet despite patterns like (2.5)-(2.8), traditional accounts remain narrowly focused on
syntactic structure. As a consequence, they overlook the semantic nature of voice alternations,
especially regarding differences among semantic event types. Anticausatives are often left
unconsidered, labeled ‘intransitive’, or treated as deviant because they do not conform to
expectation.32 But to be clear, such expectations are based on how voice has been incorrectly
described as a primarily morphosyntactic phenomenon without necessary regard for semantic
event structure. In syntactic views, voice is predicated on changes in argument structure,
especially the semantic role of the subject in relation to the verb, a quality that pertains more to
English active-passive alternations than it does to distinctions in Greek. What anticausatives in
(2.3)-(2.8) illustrate is that not all voice oppositions rely on altering grammatical relations. The
contrast between ἐγείρω ‘lift s.t. up’ vs. ἐγείρομαι (ἠγέρθην) ‘get up’ or κατακλίνω ‘lay s.t. down’
vs. κατεκλίθην ‘recline’ is not a shift in the semantic role of the subject (both active and non32

When traditional approaches describe -(θ)η- as uniquely passive in function, this shapes not just what
they do with those instances that fit the ‘rules’ but it shapes what they do with those that do not. Anticausative
alternations like the following are typically labeled ‘intransitive’ or ‘passive in active sense’: ἐκαύθη ‘burn, catch
fire’ vs. καίω ‘set on fire, cause to burn’, ἐξεπλάγη ‘be amazed’ vs. ἐκπλήσσω ‘overwhelm, cause amazement’,
ἐνεδυναμώθη ‘become strong’ vs. ἐνδυναμόω ‘strengthen’, and ἠγριώθη ‘become wild’ vs. ἀγριόω ‘make wild’.
Others like ἐκοιμήθη ‘fall asleep’ (vs. κοιμάω ‘put to sleep’) and ἐφοβήθη ‘fear, be afraid’ (φοβέω ‘frighten’) receive
‘pseudo’ labels: ‘pseudo reflexive’ and ‘pseudo passive’ respectively (Rijksbaron 2006, 160-61). While ‘pseudo’
and other alternate labels offer the ability to describe how each lexeme departs from the passive function, they do
not illustrate the pervasive semantic relationships that exist among -(θ)η-marked verbs, nor do they capture the ways
in which -(θ)η- morphology relates to the rest of the voice system.

38

active voices express the subject as agent of the action) but a shift in the semantic nature of the
event itself. A semantically causal event contrasts with a non-causal alternation, such that a
change of state may be construed either with or without external force.
Evidence of this kind of behavior, in which a single form plays more than one function,
demonstrates a shortcoming among syntactic treatments. Their attempt to separate the passive
situation, especially -(θ)η- morphology, as a distinct alternation apart from other construals, takes
no account of patterns like (2.3) and (2.4), in which a single form expresses both passive and
anticausative functions. Indeed, for languages like Greek, there is no grammatical grounding for
treating passives as distinct from anticausatives. Recall their shared attributes: In both, (1) the
object of the transitive becomes the subject of the non-active construal;33 (2) the primary figure
goes through a process of change; (3) the role of any outside force is defocused in the process.
For anticausatives, the outside force is removed and no longer present in the conception. For
passives, it is conceptually present but need not be overtly expressed. And (4), their shared
semantic basis motivates their shared formal coding. Both perfective -(θ)η- and imperfective -μαι
paradigms express passives and anticausatives with the same morphological marking. Finally,
subsuming the passive and anticausative types under a single formal expression is a relatively
common feature among languages. In fact, anticausatives often form the diachronic basis for
development of passive expressions, including Greek -(θ)η- morphology (see §3.2) (Haspelmath
1987, 1990; Croft 2012, 253; García Ramón 2014; Kulikov 2010).
To reiterate, voice alternations in Greek and other languages are not based on purely
morphosyntactic correlations but are highly dependent on lexical meaning (Croft et al. 1987;

33

In more semantic terms, the secondary participant that undergoes the action in the active becomes the
primary figure (expressed as subject) in the non-active alternation.

39

Klaiman 1991, 104). Some verbs are more agent-oriented and are more likely to occur as
passives; others are easily construed without external intervention and are likely to function as
anticausatives. This kind of distinction is not wholly foreign to English speakers since different
verbal classes may also be discerned in English based on lexical meaning and grammatical
behavior. It is easy to talk about change-of-state verbs like tear, shatter, and cool as occurring
without an agent: cloth tears, glass shatters, soup cools. But the same cannot be said for throw,
cut, or build, which require an agent. Thus, the anticausative alternation holds for tear, shatter,
and cool: ‘tear it’ vs. ‘it tears’, ‘shatter it’ vs. ‘it shatters’. But verbs that are more agent-oriented
must be expressed as passives and do not function as anticausatives in English: ‘it is thrown’ (*it
throws), ‘it is cut’ (*it cuts), ‘it is built’ (*it builds) (Haspelmath 1993, 93; 2016, 36).
Examining different alternations in this way helps to identify relevant semantic classes
that tend to behave in similar ways. These lexical classes do not exhaust the meaning of each
verb, but they do capture shared semantic components that distinguish their participation in
certain kinds of expressions, whether syntactic or morphological, highlighting the ways in which
lexical meaning impacts grammatical behavior (Levin 1993, 4-11). In this regard, the second
typological pattern concerns the relationship between voice morphology and lexical classes.
2.2.2

Consistency in the lexicon

Drawing attention to lexical classes provides further corroboration of the ways in which the
Greek voice system is functionally similar to other middle systems. From a traditional point of
view, it is often noted that Greek middle morphology expresses a fairly disparate set of events
(reflexives, reciprocals, anticausatives, passives). But when the same data are examined in light
of the typological characteristics of middle systems, rather than within an active-passive
framework, the Greek middle proves to be consistent with cross-linguistic patterns. This is true

40

in at least two ways; both are elaborated below: (1) languages with overt middle forms (Dutch,
Greek, Spanish, etc.) bear a compelling resemblance to one another in the kinds of event types
that receive middle expression. (2) A recurring pattern in such languages is the inclusion of nonalternating, middle-only verbs. In language after language, a set of verbs are basic to one voice
and do not alternate between voices. Often middle-only verbs are referred to as deponent or
media tantum. Their basic middle-only status has become a recognized hallmark among middle
systems typologically. Indeed, these middle-only verbs fit into expected semantic classes that are
commonly middle-marked across languages (Klaiman 1991, 44-109; Kemmer 1993, 21-23).
Typologically, Greek fits within a larger set of ‘middle-marking languages’, i.e. those
with overt middle forms. These ‘middle systems’ represent a diverse cross-section of languages,
including classical (Sanskrit, Greek, Latin), modern Indo-European (Germanic, Romance, Slavic
families) and non-IE languages, e.g. Fula (Niger-Congo, West Africa), Tamil (Dravidian, India),
Halkomelem (Salish, British Columbia) (Klaiman 1991, Kemmer 1993, Gerdts and Hukari 2006,
Maldonado 2008). Their diversity also extends to the shape of the ‘middle marker’ in each. In
some languages, e.g. Greek, Albanian (Geg, Albania), and Motuna (Papuan, Bougainville, PNG),
the middle form is bound up with other grammatical distinctions, e.g. person and number
agreement in Albanian la-hem ‘I wash myself’, la-esh ‘you wash yourself’. In others, middle
marking takes a stable form, as a verbal affix (e.g. Old Norse -sk, Bahasa Indonesian ber-), clitic,
or particle (e.g. French se, German sich) (Geniušienė 1987, 239-41, 301-6; Onishi 2000, 120-21;
Steinbach 2002, 47-48; Nedjalkov 2007a, 183-5; Manzini et al. 2009).
But when it comes to lexical classes, middle systems are strikingly similar. Middle
marking is used again and again for the same semantic types, as noted below with middle forms

41

in bold.34 Middle-only verbs (where known) are marked with ‘MT’ (media tantum) to indicate
their basic, non-alternating status. Two closely related types involve actions carried out on or
through one’s own body: grooming and motion. While grooming verbs like ἐπενδύομαι ‘dress’ or
νίπτομαι ‘wash’ serve as prominent examples in Greek grammars, motion verbs in (2.9) receive
less attention. Like ‘dress’ or ‘wash’, motion verbs also involve bodily change; an agentive
participant changes posture (stand, sit), location (come, go, flee), or orientation (turn, stretch).
(2.9)

Greek
ἐγείρομαι ‘rise, get up’
Hungarian emel-ked- ‘rise, get up’
Spanish pararse ‘stand up’
Greek
Kanuri
Somali

Greek

στέλλομαι ‘go, journey’
ἔρχομαι ‘go, come’ (MT)
πορεύομαι ‘walk, travel’ (MT)
Icelandic ferδa-st ‘travel’ (MT)
ὀρέγομαι ‘stretch, reach out’ Old Norse ganga-sk ‘go, leave’
tàn-t-în ‘stretch one’s body’
koma-sk ‘come’
jimics-o ‘stretch oneself’
Romanian se duce ‘go’

Together, grooming and motion constitute ‘body action middles’. Their similarity to
reflexives allows for reflexive marking in languages that do not have a middle form (e.g. English
dress oneself, stretch oneself). Typologically though, body actions are frequently distinguished
from reflexives, occurring as unmarked intransitives (I got up, stretched, and dressed).35 This
trend accords with their lexical value since such actions are commonly one-participant events
performed through one’s own bodily effort. Among middle systems, this lexical value, i.e.

34
Data from: Fula [Gombe dialect, Niger-Congo, West Africa] (Arnott 1970; Klaiman 1991), Latvian and
Lithuanian [Balto-Slavic] (Geniušienė 1987), Russian [Balto-Slavic] (Krasukhin 2006), Somali (Cushitic, Somalia)
(Saeed 1995), Spanish (Maldonado 2009). Croft, Shyldkrot, and Kemmer (1987) and Kemmer (1993) provide data
from Amharic (Semitic, Ethiopia), Ayacucho Quechua (Southern Quechua, Peru), Bahasa Indonesian (MalayoPolynesian, Southeast Asia), French (Romance), German (Germanic), Hungarian (Finno-Ugric), Icelandic (North
Germanic), Kanuri (Saharan, Nigeria), Latin (Italic), Lushai (Sino-Tibetan, Burma), Mohave (Hokan, Yuman,
California, US), Old Norse (North Germanic, Scandinavia), Pangwa (Niger-Congo, Tanzania), Romanian (Eastern
Romance), and Sanskrit (Indo-Iranian). See Kemmer (1993, 271-72) for a list of original language sources.
35
The overlap between certain middle classes and intransitive structures is evident. A middle-marked verb
in one language is often expressed as an unmarked intransitive in another (e.g. English turn, Greek τρέπομαι, Latin
reverto-r (Maldonado 2007, 854).

42

typical body action, regularly prompts their middle marking. Their verbal class and semantic
proximity to other middle types motivates their middle expression (Kemmer 1993, 53-57).36
For benefactives, contrasts like lend vs. borrow are common. These are often lexicalized,
as in suppletives (give vs. take; lose vs. find). In Greek and Amharic, the same is expressed via
middle alternation: δανίζω ‘lend’ vs. δανίζομαι ‘borrow’; bӓddӓrӓ ‘lend’ vs. tӓbӓddӓrӓ ‘borrow’.
(2.10) A. Quechua llamka-ku-y ‘work for oneself’ Fula
Fula
nyaml-aa-de ‘borrow’
Greek
δέχομαι ‘receive’ (MT)
Greek
ἐργάζομαι ‘work, trade, earn’ (MT)
λήμψομαι ‘acquire’ (MT)
Latin
Old Norse eigna-sk ‘acquire, claim’

’udd-o ‘attack’
tirf-o ‘rush at and seize’
βιάζομαι ‘dominate, ravish’
αἰκίζομαι ‘maltreat, torture’
populo-r ‘ravage, plunder’
aggredio-r ‘attack’

In (2.10), middle marking is used for actions that are ordinarily done for one’s own benefit, e.g.
choose, claim, and receive. The participant who initiates the event is also understood to be its
intended recipient or beneficiary. For some languages, this verbal class also serves as a semantic
springboard to more violent actions, extending the middle form to lexemes that express power
dynamics. The subject, as a conquering force, benefits from their superiority over a conquered
foe, e.g. σίνομαι ‘plunder’, χειρόομαι ‘overpower’ (Croft et al. 1987; Kemmer 1993, 78-81).
Reciprocal actions are also prevalent. These are semantically collective or mutual, often
involving shared affection (German sich grüßen ‘greet, welcome’; Latvian skūpstītie-s ‘kiss’) or
joint animosity, as in (2.11). In some languages, verbs of this class allow reciprocal marking (e.g.
They hugged each other). But they often occur as unmarked intransitives (cf. body actions) (e.g.

36

Alternations of this type are often expressed via lexical means in English: (a) labile alternation (e.g. the
same verb expresses both transitive and intransitive action: he turns [x] vs. [x] turns), (b) stem alternation (e.g. raise
vs. rise), or (c) suppletive alternation (e.g. bring vs. come, send vs. go). For Greek, many of the same alternations
are expressed via voice: τρέπω ‘turn [x]’ vs. τρέπομαι ‘[x] turns’, στέλλω ‘send’ vs. στέλλομαι ‘go’.

43

They hugged) since their lexical semantics are normally, or frequently, interpreted as reciprocal
in nature without the need for explicit marking.37
(2.11) B. Indonesia ber-gumul ‘wrestle’
Hungarian vere-ked-ni ‘fight’
German
sich duellieren ‘duel’ (MT) Latin
conflicto-r ‘fight’
Greek
μάχομαι ‘fight, dispute’ (MT) Russian
borot’-sja ‘fight’ (MT)
While some reciprocals alternate with non-reciprocal counterparts (e.g. Norwegian slå-ss
‘fight’ vs. slå ‘hit’), many are simply semantically collective and are thus expressed as basic
middle-only verbs (no active non-reciprocal counterpart), in keeping with their lexical value:
Latin alterco-r ‘wrangle’; Greek διαλέγομαι ‘discuss, argue’, διαπληκτίζομαι ‘exchange blows’
(Kemmer 1993, 104-11; Nedjalkov and Geniušienė 2007, 413).
Beyond physical contact, middle marking also extends to cognitive and communicative
interaction. Verbs in these classes constitute experiencer-based events; an animate participant
undergoes or experiences some kind of mental, emotional, or sensory perception. Cognition
middles involve acts of thinking, pondering, and the like. Some languages also include verbs of
intent and expectation (e.g. βούλομαι ‘wish, intend, resolve’ (MT); niyeys-o ‘intend, desire’).
(2.12) Greek διαλογίζομαι ‘consider, ponder’ (MT) Old Norse
συντίθημαι ‘agree on, decide’
Sanskrit
Latin medito-r ‘ponder, meditate’ (MT)
Somali
Latvian cīstie-s ‘decide’ (MT)

ætla-sk ‘intend’
manyat-e ‘think, believe’ (MT)
tash-o ‘consider, think over’
niyeys-o ‘intend, desire’

Speech acts describe acts of communication; a message or idea is transferred with one’s
voice (Kemmer 1993, 134). Some express reported speech; participants interact in dialogue (akin
to reciprocals). Verbs like πυνθάνομαι ‘ask, inquire’ (MT) and ἀποκρίνομαι ‘answer, reply’ (MT),
involve meanings that require a response in return or are themselves a response (Nedjalkov

37

In English, non-reciprocal verbs (poke, bite) require explicit pronominal marking in order to express
reciprocal action. Additionally, optional reciprocal marking may be used with naturally reciprocal events to stress
the reciprocality of an action in contrastive contexts: In the midst of a crowded room, my parents hugged each other.

44

2007b, 319). Others specify mental attitude or communicative intent: καυχάομαι ‘boast’ (MT),
ψεύδομαι ‘lie, speak falsely’ (MT), Lushai in-fak ‘boast’, Ayacucho Quechua llulla-ku-y ‘lie’.
(2.13) Fula tor-o ‘beg for’
Greek δέομαι ‘ask, beg’ (MT)
ἐξομολογοῦμαί ‘confess’
προσεύχομαι ‘pray’ (MT)

Lithuanian
Mohave
Old Icelandic
Pangwa

teirauti-s ‘inquire’ (MT)
mat kuna:v ‘confess’
beiða-sk ‘ask’
i-funya ‘pray’

For a full survey of cross-linguistic middle types, see §2.3. The present data may be brief
but is still representative of some common semantic classes that occur in middle systems. In this
regard, a number of observations may be made that apply to middle systems generally. First,
middle morphology is semantically consistent; middle forms are used with the same lexical
classes across a variety of languages. To clarify, this is not meant to suggest that middle-marked
verbs in each language are precisely the same nor that their formal expression is predictable
based on their meaning. Rather the semantic assertion here is that middle-marked verbs converge
around similar semantic types, with the occurrence of the middle form conventionalized in each
language. There tends to be a semantic core based on verbal class and middle-marked verbs are
created by analogy to already conventionalized lexemes (Croft et al. 1987; Haspelmath 1987).
Second, middle-only verbs are a common trait among middle-marking languages, and
they correlate with the same semantic event types most associated with middle usage (Kemmer
1993, 23; Allan 2003, 51). It is not all strange to find middle-only verbs expressing common
middle types, like typical body actions (ἀφικνέομαι ‘arrive’, οἴχομαι ‘depart’, ἅλλομαι ‘jump’;
Fula ᾽oppin-o ‘squat’), or events in which the initiator is a recipient/beneficiary of the action
(δράσσομαι ‘catch, grasp’, δέχομαι ‘receive, take’; Latin apisco-r ‘get’). As noted, such lexical
classes form a natural semantic locus for middle morphology (Kemmer 1993, 34).
Across middle systems, the verbal lexicon is organized as follows: Lexemes may be
middle-only, active-only, or alternating. Alternating verbs represent the largest portion of the

45

lexicon. This leaves non-alternating verbs as smaller sets, typically of unequal size. In some
languages, like Fula, active-only verbs appear to be more numerous than middle-only. In others,
like Hungarian, Greek, and Latin, middle-only outnumber active-only by a sizable margin
(Klaiman 1991, 106; Kemmer 1993, 22; Baerman 2007, 5; Wackernagel 2009, 160).
Any claims about which verbs occur as active-only vs. middle-only is provisional, since
their accounting depends on lexical history, diachronic variation in morphological expression,
and various patterns within the lexicon that do not apply systematically across all verbs of one
class. But some patterns do still emerge. Among active-only verbs, there are many that lack an
element of change in their lexical meaning. Verbs that express simple states are non-dynamic
and do not involve change over time, e.g. ἀσθενέω ‘be ill, weak’, νοσέω ‘be sick’. Other activeonly verbs involve body motion (like middles in (2.9)), but instead of focusing on a change of
state/location, they concern the manner of motion. The element of change is somewhat
backgrounded in their lexical meaning. These active-only verbs are atelic (durative), i.e. they do
not involve an inherent endpoint with a change of state/location, e.g. πηδάω ‘jump’, ἕρπω ‘creep,
crawl’, πλέω ‘sail, float’, βαδίζω ‘walk’, τρέχω ‘run’, and νέω ‘swim’. Conversely, many middleonly verbs of motion tend to express an inherent endpoint in their lexical meaning, e.g. ἔρχομαι
‘go, come’, ἱκνέομαι ‘reach, arrive’, οἴχομαι ‘depart, go away’, νέομαι ‘return’, χάζομαι ‘draw
back’. There are certainly exceptions to this, with manner of motion verbs (e.g. ἅλλομαι ‘jump’,
πέτομαι ‘fly’) serving as middle-only expressions as well (Allan 2003, 245). What is in focus
here is that middle-only and active-only phenomena remain a persistent, if not universal, feature
among middle systems, suggesting that both the presence and meaning of middle-only verbs in
Greek is fully consonant with typological norms (Geniušienė 1987, 299; Kulikov 2013, 275).

46

A point of clarification is warranted here in regard to how middle-only phenomena are
traditionally treated in Greek grammar. Most often, deponency is used to identify morphological
mismatches, in which the form of a verb is considered incongruent with its function. Deponents
are defined as a lexically-specified set of verbs that are ‘middle (or passive) in form’, but are
analyzed as ‘active in meaning’ based on two criteria: (a) they do not alternate between voices,
i.e. they are middle-only – expressed in middle morphology without an active morphological
counterpart, and (b) they do not express whatever is deemed to be ‘middle force’ according to
traditional views, and are thus seen as ‘in conflict’ with the rest of the voice system.
Since the majority of verbs in Greek alternate between active and non-active forms, this
is treated as a canonical pattern.38 Deponents are then viewed as exceptions to the rule, in which
the form of a verb (middle) does not match its function (active). Without a certain ‘middle
meaning’ a middle-only verb is seen as playing an active role in the voice system. Middle-only
verbs like ἔρχομαι ‘go, come’, ἐργάζομαι ‘work, trade, earn’, and βούλομαι ‘wish, intend, prefer’
are often included in lists of deponents as ‘active in meaning’ (Smyth 1956, 90; Wallace 1996,
428-30; Baerman 2007, 2-7; Lavidas and Papangeli 2007, 97; Ladewig 2010, 135-6).39
A separate examination is then required to explain the presence of these exceptional
cases. Yet traditional accounts offer no principled basis for their existence, primarily relying on

38
The descriptive focus here tends to be on establishing morphosyntactic oppositions in the voice system.
A formal opposition in voice morphology is described in terms of a corresponding functional opposition in syntactic
structure. For a verb like ποιέω ‘make’ vs. ποιέομαι ‘make for oneself’, its morphological contrast is linked to a
parallel functional contrast in the role of the clausal subject. Traditionally, the middle is said to express more subject
focus, involvement, or participation in the action (Wallace 1996, 414). Once a canonical pattern like this is
confirmed for the majority of verbs, it provides a means of identifying deviations from the norm. Thus, the central
concern becomes distinguishing between normal and exceptional behavior. If the normal realization of voice
morphology is tied to an oppositional contrast between alternating middles and their active counterparts, then any
lexemes that depart from this pattern are considered non-standard in some way (Baerman 2007, 4-6; Corbett 2007).
39
The intuition about what is considered ‘active in meaning’ relies heavily on English syntax (see §2.1) to
the exclusion of typological norms among middle systems.

47

ad hoc rules or arbitrary lists of verbs that change with each account (Wallace 1996, Ladewig
2010). For example, Lavidas and Papangeli (2007) engage with, but ultimately reject semantic
patterns in the Greek middle. An overreliance on morphosyntactic features leaves them illequipped to deal with the kinds of idiosyncrasies that naturally arise with lexical behavior (see
3.1 for discussion). This natural idiosyncrasy in lexical semantics then becomes an argument
against semantic approaches rather than a case against their own syntactico-centric explanation.
Their focus remains on transitive deponents, defined as verbs that bear middle/passive
morphology in transitive syntax, i.e. they occur with an accusative object. Lists of deponents
include middle-only verbs like ἀποκρίνομαι ‘answer’ (illustrated among speech act middles in
(2.13) above), as well as μάχομαι ‘fight’ (listed among reciprocal middles in (2.11)). According
to this view, their middle marking is not semantically motivated but is instead the result of a
morphological feature specification that requires them to be marked with a pre-specified form
(i.e. middle) even though they appear in what is defined as active syntax. This idiosyncratic rule
applies at random across the lexicon to create the familiar mismatch between morphology and
syntax that defines traditional notions of deponency (120).
Yet analyses of this sort only defer explanation to another level. To explain the presence
of deponent phenomena, a random morphological rule is posited. But no clarification is given for
why such an anomalous rule should exist in the first place. Nor is there any elucidation for how a
morphological rule could apply at random in each language and still produce a semantically
consistent set of verbs both language-internally in Greek and cross-linguistically among middlemarking languages. No principled explanation is offered for why certain event types consistently
appear as middle-only verbs in Greek and across middle systems. Such analyses cannot account
for why certain semantic types like translational motion (self-propelled motion along a path:

48

ἔρχομαι ‘go, come’), naturally reciprocal actions (μάχομαι ‘fight’), cognition/expectation (οἴομαι
‘think, suppose, expect’), and speech acts (ὀλοφυρομαι ‘lament’) consistently appear as middleonly verbs across languages while other semantic types like creation/destruction (οἰκοδομέω
‘build’, καταλύω ‘tear down’) or physical impact (σχίζω ‘tear apart’, νύσσω ‘pierce’) and killing
(ἀποκτείνω ‘kill’) do not (Kemmer 1993, 33, 106; Beavers 2011).
One of the consequences of traditional syntactic accounts is that cross-linguistic
consistency gets treated as though it is language-specific anomaly, i.e. a problem to be solved.
But, if we pay attention to typological regularities, then the status of middle-only verbs as
deviant or exceptional must be called into question. Their typological persistence alone
demonstrates that they ought to be treated as a general characteristic of such languages,
warranting their place in an overall explanation of normal behavior among middle systems.
A more general account that gives credence to their cross-linguistic distribution is
preferable to one that treats them as incompatible with regular middle patterns (Klaiman 1991,
44, 57-58; Kemmer 1993, 23-34; Matthews 2007; Maldonado 2009). The tendency for languages
to mark the same kinds of event types with middle morphology illustrates how Greek voice has
less to do with morphosyntactic oppositions and more to do with lexical classes. Deponent, or
middle-only verbs, fit into event classes that are semantically motivated within the voice system,
and represent a natural consequence, or by-product, of the semantic nature of middle marking.
Greek voice phenomena are better understood as the grammatical categorization of
similar semantic types. Lexical classes are organized together with a particular marking pattern
to express their shared conceptual basis. Middle morphology is extended verbs with suitable
semantic characteristics regardless of the presence of an active counterpart (Kemmer 1993, 33).
Section 2.3 considers these classes in more detail, with a full survey of common middle types.

49

This provides a broader comprehension of how Greek compares and contrasts with other middle
systems and paints a wider picture of how middle voice behaves across the world’s languages.
2.3 Inventory of middle types
This section provides a typological survey of middle event types. Geniušienė (1987) and
Kemmer (1993) supply cross-linguistic data; this is combined with language-specific studies as
well.40 Following Kemmer, I include data from languages that are genetically and geographically
diverse. I also follow her groupings in regard to various cross-linguistic middle types.41 As in
§2.2.2, middle forms are in bold to indicate where each lexeme receives middle marking.
Languages may use different formal means to express the middle, yet they show significant
semantic overlap in the types of events they include in the middle domain.
One of the most recognizable middle types is grooming/body care. An animate entity
performs an action that affects one’s body, body parts, or inalienable possessions. Such events
include routine maintenance of the body as well as culturally specific ritualized actions, such as
anointing with oil or wearing a veil (Kemmer 1993, 54; Rijksbaron 2006, 144; Allan 2003, 89).
(2.14) Bella Coola
French
German
Greek
40

sx̌=aax̌uc-m ‘shave beard’

Kanuri
s’-habiller ‘dress, wear’ (MT) Latvian
sich anziehen ‘dress’
Mohawk
κατακαλύπτομαι ‘veil oneself’ Somali

hàr-t-în ‘wash’ (MT)
apsegtie-s ‘cover oneself’
-ate-nawirohar- ‘brush one’s teeth’
dhay-o ‘rub oneself with’

Language data: Balinese [Malayo-Polynesian, Bali] (Shibatani and Artawa 2007), Bella Coola [Salish,
British Columbia] (Beck 2000), Danish [N Germanic] (Sansò 2006, 237), German (Steinbach 2002), Halkomelem
[Salish, British Columbia] (Gerdts and Hukari 2006), Latvian and Lithuanian [Balto-Slavic] (Geniušienė 1987),
Mohawk [Iroquoian, northeastern North America] (Mithun 2006, 201), Motuna [Papuan, Papua New Guinea]
(Onishi 2000), Muscogee [Muskogean, Oklahoma, US] (Martin 2000, 381-87), Otomi [Oto-Pamean, Mexico]
(Palancar 2004), Russian [Balto-Slavic] (Krasukhin 2006), Somali (Cushitic, Somalia) (Saeed 1995), Spanish
(Maldonado 2008, 2009). From Kemmer (1993): Ayacucho Quechua (S Quechua, Peru), Bahasa Indonesian
(Malayo-Polynesian, SE Asia), French (Romance), Fula (Gombe, Niger-Congo, W Africa), Hungarian (FinnoUgric), Icelandic (N Germanic), Kanuri (Saharan, Nigeria), Latin (Italic), Lingala (Niger-Congo, Bantu), Mohave
(Hokan, Yuman, California), Old Norse (N Germanic, Scandinavia), Russian (Balto-Slavic), Sanskrit (Indo-Iranian),
Tsonga (Niger-Congo, SE Bantu), and Turkish (Oghuz, Turkey).
41
Middle types in Kemmer (1993): grooming, motion (e.g. translational, non-translational, change in
posture, and collective), indirect/self-benefactive, naturally reciprocal, mental (emotion, cognition, perception,
speech act), spontaneous events, logophoric, and passive/facilitative/impersonal middles. Each is discussed below.

50

Benefactives in (2.15) are performed on a second figure so there is a direct object in the
clause, but the agent also plays the role of recipient or beneficiary of the action (Kemmer 1993,
78; Rijksbaron 2006, 147-50; Allan 2003, 112).
(2.15) Greek

κομίζομαι ‘get back, recover’ Sanskrit
περιποιέομαι ‘obtain’
Spanish
ἐκλέγομαι ‘choose’ (MT)
Latin
potio-r ‘get possession of’
Tsonga
Latvian izlasītie-s ‘choose for oneself’ Turkish

labhat-e ‘obtain, receive’
conseguirse ‘get’
allegarse ‘obtain’
ku ti-phahlela ‘offer sacrifice for self’
ed-in- ‘acquire’

Similar to body care in (2.14), there are a number of types that involve various facets of
body action or motion carried out through the body. Self-induced actions in (2.16) involve a
participant changing location along a path through space (Talmy 1985; Kemmer 1993, 57).
(2.16) B. Indonesia
Balinese
Fula
Greek

ber-djalan ‘walk, stroll’
ma-laib ‘run’
ma’y’y-o ‘climb, mount’
ἅλλομαι ‘leap, jump’ (MT)
πέτομαι ‘fly’ (MT)
ἀνάγομαι ‘set sail’
Halkomelem pəpətén̓əm̓ ‘sailing’

Halkomelem
Lithuanian
Mizo
Muscogee
Otomi
Somali
Spanish

cƛə̓ m ‘jump’
bastyti-s ‘wander, roam’ (MT)
in-shon ‘move’ (MT)
ta·sk-itá ‘jump (of one)’
n-tsoni ‘fly’
noq-o ‘return’
regresarse ‘return’

These translational motions are contrasted with non-translational motions. While events
in (2.17) do not necessarily express a change in location, they do involve bodily movement that
alters the body’s (or body part’s) configuration (Kemmer 1993, 57; Allan 2003, 77).
(2.17) Greek

ἐπεκτείνομαι ‘stretch out’
στρέφομαι ‘turn oneself’
Halkomelem c̓aləsəm ‘turn around’
Icelandic
snua-st ‘turn’

Lingala
Lithuanian
Old Norse
Spanish

bóngwa-na ‘turn around’
versti-s ‘turn (side to side)’
hrista-sk ‘shake (the head)’
estirarse ‘stretch out’

Change in posture in (2.18) is similar to non-translational motion, but here the body ends
up in a different spatial plane, e.g. Somali foorors-o ‘bend over, stoop’ (Kemmer 1993, 56).
(2.18) A. Quechua
Balinese
German
Greek

tiya-ku-y ‘sit down’
ma-tangi ‘stand up’
sich hinlegen ‘lie down’
κατάκειμαι ‘lie down’

Lithuanian
Muscogee
Otomi
Sanskrit

leñkti-s ‘bend’
wakk-itá ‘lie (of one)’
n-dondi-hmu ‘kneel’
nipadyat-e ‘lie down’

51

Reciprocals denote some level of mutual engagement for event participants. Alternations
express a contrast between symmetrical vs. asymmetrical action. Either participants act on one
another with symmetrical force or one participant acts on a second, with one direction of force,
e.g. Icelandic spyrja-sk ‘ask one another’ vs. spyrja ‘ask’ (Croft 2012, 243). Such verbs often fall
into a number of types: spatial relations (join, associate), identity or likeness (be similar, be like),
and human relations (meet, marry, make a covenant) (Kemmer 1993, 117; Nedjalkov 2007c, 96).
(2.19) Balinese
German
Greek
Kanuri
Latin

ma-palu ‘fight each other’
sich fraternisieren ‘fraternize’
συναγωνίζομαι ‘fight, contend with’
tòkùm-t-âi ‘copulate [of insects]’(MT)
amplecto-r ‘embrace’
osculo-r ‘kiss’

Latvian
Lingala
Old Norse
Otomi
Somali
Tsonga

sastaptie-s ‘meet each other’
-kamata-na ‘resemble’
hitta-sk ‘meet’
m-phɔts’i ‘help each other’
dhaxso ‘get married’
ku ringa-na ‘be similar’

Collective events also involve multiple participants and may be a subtype of (2.19) since
they too involve group actions. Individuals act collectively. The group is conceived as a whole
carrying out a single action together (Kemmer 1993, 123-25).
(2.20) German
Greek

sich sammeln ‘gather, assemble’
συνέρχομαι ‘gather together’
ἀθροίζομαι ‘gather together’
Hungarian tüle-ked- ‘throng’
Latin
misceo-r ‘assemble, unite’

Latvian
Lithuanian
Old Norse
Russian
Somali

pulcētie-s ‘come together’
eiliuoti-s ‘line up’
flykkja-sk ‘flock together’
sobrat’-sja ‘collect, gather’
yoobso ‘assemble, gather’

Among experiencer-based events, middle marking is used to express cognition, emotion,
perception, and speech. In the cognitive domain, mental activities express mental effort on the
part of the experiencer subject, including thinking, calculation, and conjecture.
(2.21) French s’aviser (de) ‘think up’ (MT) Hungarian gondola-kod- ‘think, muse, meditate’
Fula
hiis-o ‘calculate’
Latin
comminīsco-r ‘think, devise’
Greek τεχνάομαι ‘contrive, craft’ Old Norse pykkja-sk ‘think (that), opine’
Emotion middles involve emotional reactions of one sort or another, often expressing a
lower degree of volitionality than mental activities. These emotional/mental changes tend to
involve less control for the subject participant (Kemmer 1993, 128-35; Allan 2003, 64).

52

(2.22) French se réjouir ‘rejoice’ (MT)
Greek μαίνομαι ‘become mad’ (MT)
μιμνῄσκομαι ‘remember’
ὀργίζομαι ‘become angry’
φοβέομαι ‘fear, be afraid’
εὐφραίνομαι ‘rejoice’
Latvian bītie-s ‘fear’ (MT)

Lingala
Mohave
Otomi
Sanskrit
Spanish
Somali

-tataba-na ‘be troubled, astonished’
mat iθa:v ‘be angry’ (MT)
n-sente ‘be/get sad’
ramat-e ‘rejoice’
acordarse ‘remember’
garwaaqs-o ‘recall, remember’
cabs-o ‘fear, be afraid’

For speech act verbs, an internal reaction is expressed as outward speech, including both
positive (e.g. ἀσπάζομαι ‘welcome, greet warmly’) and negative responses (e.g. μέμφομαι ‘blame,
censure’; Spanish quejarse ‘complain’) (Kemmer 1993, 133).
(2.23) Balinese
Bella Coola
French
Fula
Greek

ma-takon ‘ask’
ckʷ= uc-m ‘start talking’
se plaindre ‘complain’
wull-o ‘lodge complaint against’
μωμάομαι ‘criticize’ (MT)

Greek
Halkomelem
Hungarian
Latin
Spanish

ἐντέλλομαι ‘command’
kʷecəm ‘scream, yell’
dicse-ked- ‘boast’
preco-r ‘ask for’ (MT)
jactarse ‘brag’ (MT)

Perceptions describe sensory experiences. An animate participant perceives a state of
affairs through the senses, e.g. taste, touch, smell (Kemmer 1993, 136; Allan 2003, 95).

̓ m ‘look around’
(2.24) Bella Coola kxGreek

Latin

conspicio-r ‘perceive’ (MT)

ὀσφραίνομαι ‘smell’ (MT)
odōro-r ‘smell’ (MT)
περιβλέπομαι ‘look around’ Lithuanian dairyti-s ‘look around’
γεύομαι ‘taste’ (MT)
Somali
dhadhanso ‘taste s.t. for oneself’

For types above, the middle involves an animate participant who does or experiences
something through mind or body, but for the following types this isn’t necessarily the case.
Terms like ‘spontaneous process’/‘change-of-state’ are used widely in typological literature to
describe events that are non-agentive; a physical change of state or process of change occurs
spontaneously without explicit involvement of an agentive or external force (Kemmer 1993,
2003; Shibatani 2006). Events in (2.25) can involve animate subjects (wake up, fall asleep), but
many express biological/physiological processes that occur with inanimate entities (ξηραίνομαι
‘dry, wither’; αὐξάνομαι ‘grow’), existential changes (γίνομαι ‘come about, happen’; φαίνομαι
‘appear’), and changes in attribute/property (λευκαίνομαι ‘become white’).

53

Some languages include involuntary movement, e.g. Latin temblo-r ‘tremble’, Sanskrit
vepat-e ‘tremble, shiver’, Lingala –ninga-na ‘move, tremble’, and Greek σείομαι ‘shake’. Greek
σείομαι can be used with animate participants (e.g. guards tremble in Matt. 28:4) and inanimate
entities (e.g. the earth shakes in Matt. 27:51) (Kemmer 1993, 142-45).
(2.25) Balinese

burqo ‘pour out’
buko ‘become sick’
Bella Coola x̌up-m ‘sink (in mud)’
Greek
ἐκχέομαι ‘pour out, spill’
ἐλαττόομαι ‘become less’
ὀπτάνομαι ‘appear’
Halkomelem liqʷəm ‘get calm’ (weather)

Lithuanian lieti-s ‘stream, pour’
Old Norse gróa-sk ‘grow’
søkkva-sk ‘sink’
Sanskrit
várdhate ‘grow’
Spanish
dormirse ‘fall asleep’
aparecerse ‘appear’
Turkish
gör-ün ‘appear’

Examples in (2.26) illustrate a number of semantically similar types that commonly
receive middle expression, e.g. passives, facilitatives, and impersonals. Passive events involve
two participants. The agent (as external cause that initiates the action) is given a lower degree of
relevance compared with the patient, the participant that undergoes the effects of the action. In a
typical passive, the patient is promoted to subject status as primary figure in the event (be hit by).
The agent is either left unexpressed or is demoted to an oblique phrase, reflecting its status as
pragmatically deemphasized (Haspelmath 1987, 29; Kemmer 1993, 147; Shibatani 2006, 248).
In the facilitative (or potential passive) type, a semantic agent is implied in the event, but
it cannot be overtly expressed, e.g. French le livre se vend bien ‘the book sells well’. Similar
expressions occur in German and Modern Greek (Steinbach 2002, 37, 48; Haspelmath 1987, 7).
A quality judgment is asserted about the relative ease or difficulty by which an event occurs.
Similar cases occur in Kanuri, expressing whether or not something is edible/potable. The event
proceeds from the patient based on an inherent quality, allowing the event to progress or not. In
the impersonal middle type, the agent is generic and is left unexpressed, as in Spanish: se habla
mucho aquí ‘there’s a lot of talking here’ (Geniušienė 1987, 262-67; Kemmer 1993, 147).

54

(2.26) Danish
German
Greek

Bogen sælge-s
Dieser Rasen mäht sich schnell
σπείρεται
κρίνομαι
Halkomelem pas-ət-əm
Kanuri
Latin
Russian
Spanish

kʷən-ət-əm
t-úrúk-ìn
t-ə́wìnbâ

amo-r
dom stroit’-sja
vender-se

‘The book is sold’
‘This lawn mows quickly’
‘be sown’42
‘be tried, be condemned’
‘be hit by’
‘be taken by’
‘I am seen’
‘it is not eaten/edible’
‘be loved’
‘the house is being built (by)’
‘be sold’

The final type is the logophoric middle in which middle marking is used in complement
clause constructions with a range of reportive verbs (believe, say, claim, and decide). The middle
is used for this type in Old Norse and Modern Icelandic. While Ancient Greek does express
similar ‘accusative-plus-infinitive’ constructions, they do not directly correlate with the use of
the middle voice form (Smyth 1956, 442, 449; Kemmer 1993, 81-93; Allan 2003, 44).
(2.27) Modern Icelandic
Haraldur segi-st
hafa
skrifað
ritgerð-ina
Harold
says-MID to.have
written
thesis-the.ACC
‘Harold says that he has written his thesis’ (Andrews 1982)
Event types in (2.14)-(2.27) represent major cross-linguistic middle uses in both IE and
non-IE languages. The functional domain of the middle shows remarkable consistency across
languages, spanning semantic types from body care and reciprocals, to body motion and
cognition events, as well as passives and passive-like semantics within the scope of a single
grammatical form. And yet even among middle-marking languages, variation persists.
Expressive flexibility allows languages to maintain alternate grammatical routes to similar
communicative paths. Greek and Halkomelem subsume the passive type within the scope of the

42

A middle form like this can be used in impersonal events in Greek such as in LXX Numbers 20:5 τόπος οὗ
οὐ σπείρεται ‘a place where there is no sowing (of seeds)’ or ‘a place of no sowing’.

55

middle form, but other languages, such as French, do not. French includes the facilitative and
impersonal uses within the middle-marking domain, but the passive function is expressed as a
separate type with a distinct grammatical form (Kemmer 1993, 149).
Further differences appear in Halkomelem. The middle form in Halkomelem fits well
within typological middle patterns, but two event classes receive middle marking (-m) that are
not attested as major cross-linguistic types, e.g. body processes and verbs of emission (Gerdts
and Hukari 2006, 53). Body processes are involuntary acts that do not result in a full change of
state. But much like Greek σείομαι ‘shake’, Halkomelem includes sneeze, cough, tremble, and
breath in this type. There are other languages that include body processes in the middle-marking
domain, including Greek (πέρδομαι ‘break wind’, χρέμπτομαι ‘spit, cough’, and χασμάομαι ‘yawn,
gape’) and Sanskrit (svedat-e ‘sweats’ and kāsat-e ‘coughs’). Such verbs demonstrate some level
of patient-like self-affectedness inherent in their semantics since they are generally not events
that can be performed on another participant (Kemmer 1993, 61).
Verbs of emission in Halkomelem describe inherent properties of participants, e.g. shine,
flicker, or smell foul. At the same time, Halkomelem excludes verbs of cognition/emotion in the
scope of the middle form. No verbs of these types receive the verbal suffix -m (Gerdts and
Hukari 2006, 50-52). Ancient Greek is similar; it too omits types that are exhibited in other
languages. Neither the logophoric nor facilitative types are expressed as middles in Greek. Yet
Classical and Hellenistic Greek fit within cross-linguistic trends in expressing all other major
types: grooming, benefactive, motion, reciprocal, cognition and emotion, speech act, perception,
spontaneous process, and passive (Kemmer 1993, 16-20; Allan 2003, 43). The middle category
may be typologically diffuse, i.e. no two middle systems are exactly alike, but there remains a
strong semantic overlap in the kinds of events that receive middle expression across languages.

56

3. Middle diachrony
The previous chapter examined the Greek middle in the context of cross-linguistic patterns in
voice. Greek middle morphology, and especially the presence of middle-only verbs, may seem
erratic or illogical if considered from the persective of an active-passive system. But when
brought into the context of voice typology, their cross-linguistic consistency attests to the
fundamentally semantic nature of middle systems. The present chapter focuses on historical
processes among middle-marking languages, particularly how middle marking and middlemarked verbs evolve and change over time, with the Greek middle taking centre stage. Historical
developments are examined through the lens of two diachronic processes, both of which concern
the intersection of voice morphology and lexical semantics. Section 3.1 considers lexicalization
and distributional idiosyncrasies that commonly occur among middle-marking languages.
Section 3.2 pertains to the development of -(θ)η- morphology in Greek, following its path from
Proto-Indo-European (PIE) to Classical and Hellenistic periods. Essentially, the -(θ)η- form rose
through a process of grammaticalization, expanding its use in the voice system from a more
lexical/derivational form (restricted/irregular) toward a more inflectional one (general/regular).
The grammatical status of morphological forms revolves around the extent to which their
behavior is idiosyncratic or systematic (Bybee 1985, 81-86). Inflectional forms tend toward
greater grammaticality, exhibiting highly general, regular, and systematic patterns. By contrast,
derivational forms tend toward greater lexicality, showing more idiosyncrasy, distinct shifts in
meaning, and less systematic patterns. These two poles (lexical vs. grammatical) provide a
continuum of behavior with the possibility of change over time. As form-meaning pairs are used
in language, their status may shift by degrees, toward one pole or the other (Kiefer 2000, 300;
Brinton and Traugott 2005, 2, 101-2).

57

Changes in grammatical status are often the result of diachronic processes like
lexicalization and grammaticalization. The former refers to semantic enrichment and the creation
of more contentful forms. When complex words or phrases take on idiosyncratic properties, both
formal and semantic, this can lead to a loss in semantic segmentability over time. The meaning
of a lexicalized element cannot be understood on the basis of its constituent parts, a phenomenon
that encompasses anything from syntactic phrases (a death-in-the-afternoon is a cocktail) to
lexical compounds (Modern English (ME) bailiwick ‘a person’s sphere of interest or skill’ <
Middle English bailiffwic ‘jurisdiction of a bailiff’ < bailie ‘bailiff’ + wick ‘town, district’
(Brinton and Traugott 2005, 95-97; Sauer 2004, 1626; Bauer 2003,44- 45).
Conversely, grammaticalization refers to a process of semantic bleaching. Elements
become more grammatical, functional, and systematic. As certain forms lose specificity of
meaning, they adopt more generalized purposes, altering their function and expanding their
distribution in the process (e.g. Old English magan ‘have strength to’ > ME may) (Brinton and
Traugott 2005, 99; Bybee 2010, 106). Section 3.2 focuses on this kind of grammatical change, as
it pertains to the perfective -(θ)η- form and its integration into Greek voice morphology.
3.1 Lexicalization in voice
Voice categories often behave in ways that are typical of derivational morphology,43 a trait
especially true of middle systems. Middle-marking languages generally subsume a host of

43
I use the term ‘derivational’ here in contrast to inflectional morphology, in order to draw attention to the
idiosyncratic distribution of the middle voice form across the verbal lexicon and its semantic behavior among lexical
classes. Derivational and inflectional morphology exist on a continuum without a sharp distinction between them
(Bybee 1985, 11-12). Inflectional affixes are typically obligatory forms that apply systematically across all verbs, as
higly productive expressions. As such, they must be combinable with any verbal stems and yield a predictable and
generalized meaning. Derivational affixes, on the other hand, are closer to lexical forms. They tend to apply in an
idiosyncratic manner across the verbal lexicon and are restricted to certain semantic classes of verbs. They derive
new semantic expressions that are more distinct in meaning from their bases. In §2.1, the term ‘derived’ refers to a
type of voice system in which a marked voice is syntactically derived from a basic unmarked expression.

58

semantic functions within the scope of the middle form, including reflexives, reciprocals, and
anticausatives, all of which are often themselves described in derivational terms (Haspelmath
1987, Nedjalkov 2007c). Two major patterns point to idiosyncrasies in voice. The first is that
middle morphology frequently displays formal idiosyncrasies in its verbal marking patterns,
resulting in unpredictable limitations on its productivity (§3.1.1). The second involves semantic
idiosyncrasies in voice and its propensity toward lexicalization (§3.1.2) (Bybee 1985, 20-21, 32;
Haspelmath and Müller-Bardey 2004; Brinton and Traugott 2005, 87, 92).
3.1.1

Formal idiosyncrasies

Examples of formal idiosyncrasy include the following: (1) verbal triads created via labile
alternations, (2) semantic doublets, and (3) diachronic shifts in voice morphology. Among
middle systems, it is common to find a degree of synchronic variation in the expression of
certain event types. In some cases, there may co-exist a parallel means of expressing the same
type of alternation (Geniušienė 1987, 253, 261; Kemmer 1993, 21). In (3.1), the anticausative
function in Greek is primarily expressed through voice alternation (Sausa 2016). A change in
voice morphology signals a grammatical contrast between causative (active) and anticausative
(middle) events. An alternate means of expressing the same contrast occurs in (3.2) with labile
alternations. Here, there is a change in meaning between alternants (causative vs. anticausative),
but no change in form. The same lexical form, i.e. basic active, expresses both alternants.
(3.1)

κινέω ‘set [x] in motion’
χωρίζω ‘separate, divide [x]’
ψύχω ‘cool [x]’

κινέομαι ‘[x] moves’
χωρίζομαι ‘[x] leaves, departs’
ψύχομαι ‘[x] cools’

(3.2)

ἀνατέλλω ‘cause [x] to rise’
ἀπορίπτω ‘drive [x] away’
ἐνισχύω ‘strengthen [x]’

ἀνατέλλω ‘[x] rises’
ἀπορίπτω ‘[x] jumps off’
ἐνισχύω ‘[x] grows strong’

With this kind of variation, it is only natural to find some lexemes that follow both
patterns, as in (3.3), resulting in verbal triads of the anticausative type. In middle systems, active

59

and middle forms of the same verb may co-exist as synchronic variants with both active and
middle forms expressing the same anticausative event (Junčytė 2018, 80-95). Between στρέφω ~
στρέφομαι ‘[x] turns’, there is a change in form, but no change in meaning.
(3.3)

Greek

στρέφω ‘turn [x]’
καθίζω ‘seat [x]’
αὐξάνω ‘cause [x] to grow’
Lithuanian virsti ‘tumble, turn [x]’
klišinti ‘make [x] slipshot’
Russian
starit’ ‘make [x] old’

στρέφω ~ στρέφομαι ‘[x] turns’
καθίζω ~ καθίζομαι ‘[x] sits’
αὐξάνω ~ αὐξάνομαι ‘[x] grows’
virsti ~ virsti-s ‘[x] tumbles, turns’
klišti ~ klišinti-s ‘[x] becomes slipshot’
staret’~ starit’-sja ‘[x] grows old’

Verbal triads in (3.3) relate to one another in one of two ways. Either they are largely
interchangeable, collocating with the same noun phrases and appearing in similar discourse
contexts, or they vary in terms of lexical senses and other relevant parameters. Some verbal
triads in Greek diverge in relation to other verbal categories; (3.4) splits according to mood, with
the active used only in the imperative and the middle form used elsewhere. In contrast, (3.4)
varies with aspect. The active is preferred in the (aorist) perfective and the middle in the
(present) imperfective (Geniušienė 1987, 106-7, 261; Allan 2003, 209).
(3.4)

(a) ἐγείρω ‘raise [x]’
(b) ἀνίστημι ‘raise [x] up’

ἔγειρε [Imp.] ‘get up!’ ~ ἐγείρομαι ‘[x] rises’
ἀνέστην [Aor.] ‘[x] stood’~ ἀνίσταμαι [Pres.] ‘[x] stands’44

Rather than showing systematic patterns across verbal paradigms, middle marking tends
to be more idiosyncratic in its distribution and is often contingent on lexical semantics and the
history of a given verb. This leads to the second type of formal idiosyncrasy. Middle systems
44

These kinds of patterns are often shaped by historical factors including lexical value, formal analogy in
morphological expression, and historical shifts in meaning and usage. A possible formation process for ἀνίστημι
(derivative of ἵστημι) may start with an intransitive active aorist ἔστην ‘stood’ in Greek (compare with Sanskrit
ásthām < PIE h2é-steh2-m) corresponding to other active forms (e.g. ἔφυν ‘grew’). Greek creates a causative active
form, resulting in aorist causative/anticausative pairs: ἔστην ‘stood’ vs. ἔστησα ‘caused to stand’, ἔφυν ‘grew’ vs.
ἔφυσα ‘caused to grow’. By analogy, a causative present develops (ἵστημι). Or, an already existing active intransitive
present ἵστημι ‘stand’ takes on a causative meaning ‘cause to stand’. Finally, an anticausative middle develops in
opposition to the causative present, yielding ἵστημι ‘cause to stand’ vs. ἵσταμαι ‘stand, stand up’ (Allan 2003,
209n362; Beekes 2010, 601). In this history, formal coding develops via analogy and oppositional contrast, an active
intransitive produces a causative alternation, which then provides the impetus for an anticausative middle.

60

often involve some level of idiosyncratic distribution among lexical types. Speech acts like
ἀποδύρομαι ‘lament’, ὀλοφύρομαι ‘wail over’, and μωμάομαι ‘criticize’ consistently receive
middle marking, whereas similar verbs like οἰμώζω ‘wail’ κλαίω ‘weep’, and γογγύζω ‘grumble’
do not. Likewise, in German, sich hinsetzen ‘sit down’ and sich hinlegen ‘lie down’ are middle
marked, but aufstehen ‘stand up’ is not. And in Tsonga, ku ti-stakela ‘be/become happy’ is
middle, but ku hundzuka ‘be/become angry’ is not. Which verbs receive middle expression, and
which do not must be lexically specified; the event type of a given verb does not necessarily
guarantee its voice morphology (Kemmer 1993, 21; Haspelmath and Müller-Bardey 2004, 1139).
Such idiosyncrasy may further result in the creation of semantic doublets or overlapping
distribution of active and middle forms for the same verb. In a semantic class, there may be some
variation in voice morphology, such that either form is possible (cf. middle verbs of asking:
δέομαι ‘beg, plead, ask’, πυνθάνομαι ‘inquire, ask’ vs. active verbs of the same: ἐρωτάω ‘ask,
request’). This prompts the rise of synonymous pairs (e.g. αἰτέω ~ αἰτέομαι ‘ask, request, beg’),
in which active and middle forms serve as synchronic variants with little to no difference in
meaning. Semantic doublets are widely attested among middle systems: Czech vzpomínat ~
vzpomínat si ‘recollect’, German irren ~ sich irren ‘err’, Greek πειράω ~ πειράομαι ‘try,
experience’, Hungarian kéredz ~ kéredz-ked- ‘ask, request’, Norwegian skimle ~ skimle-s ‘grow
mouldy’, Lithuanian bijóti ~ bijóti-s ‘fear’, Spanish engordar ~ engordar-se ‘grow fat’ and
Swedish rosta ~ rosta-s ‘become rusty’ (Genuišienė 1987, 137, 282-90; Krasukhin 2006, 91).
Finally, the third type of formal idiosyncrasy arises through diachronic shifts in voice
morphology, as in (3.5). Classical Greek usage patterns show an alternation between active and
middle. But in Hellenistic, the active has largely fallen out of use, leaving the middle to remain.
This results in a media tantum expression: A middle form with no active counterpart available.

61

(3.5)

Classical
(ἀποκρίνω ‘separate, distinguish’)
(πορεύω ‘make go’)
(ορχέω ‘make dance’)

Hellenistic
ἀποκρίνομαι ‘reply to a question’
πορεύομαι ‘go’
ὀρχέομαι ‘dance’

Frequency effects can shape which elements endure and which do not. Token frequency
(i.e. number of instances of a lexeme) strengthens lexical elements. Those with a sufficiently
high frequency will be stronger than those with low frequency. This lexical strength has two
consequences for middle verbs in (3.5). High-frequency lexemes are (1) more readily accessible
for decision-making in language production. And (2), they show increased lexical durability over
time, with a lower likelihood of undergoing analogical change. That is, token frequency has a
conserving effect. Middle verbs in (3.5) are more likely to remain middle despite a loss in voice
alternation (Allan 2003, 55; Brinton and Traugott 2005, 17; Bybee 2007, 280; 2010, 24, 75).
Among other verbs, however, this conserving effect is not present. One lexeme that
undergoes analogical change is εὐαγγελίζω ~ εὐαγγελίζομαι ‘announce good news’. The middle is
the older formation, with the active serving as a later Hellenistic variant. In this case, the active
form may arise by formal and semantic analogy to other verbs of its kind: διαφημίζω ‘spread
news abroad, report’ and ἐμφανίζω ‘make a report, inform’. This pattern follows the expansion of
-ιζω morphology in the Hellenistic period (e.g. older Classical forms like ῥαίνω take on
Hellenistic variants like ῥαντίζω ‘sprinkle’).
3.1.2

Semantic idiosyncrasies

Semantic idiosyncrasies are also common among middle systems. Alternations in (3.6) are
formally relatable in their morphological alternation, but semantically distinct and idiosyncratic
in their meaning compared to other active-middle alternations. Instead of expressing a general or
regular shift in meaning, their semantic relation has become essentially lexicalized, reflecting the

62

natural semantic idiosyncrasies that arise through lexical development (Bybee 1985, 21;
Geniušienė 1987, 136, 300; Haspelmath 1987, 14; Wiemer and Nedjalkov 2007, 468).
(3.6)

Bulgarian kanja ‘invite’
Czech
hádati ‘tell fortunes’
hoditi ‘throw, fling’
French
douter ‘doubt’
German
unterhalten ‘to entertain’
zerwerfen ‘to annul’
Greek
ἅπτω ‘set on fire, kindle’
γαμέω ‘marry (of men)’
συμβάλλω ‘compare, wage war’
Icelandic taka ‘take’
Latvian
zāgt ‘steal’
Norwegian syne ‘show’
Spanish
volver ‘to turn, return’
Swedish finna ‘find’

kanja se ‘intend’
hádati se ‘quarrel, argue’
hoditi se ‘be suitable’
se douter ‘suspect, conjecture’
sich unterhalten ‘to talk, chat’
sich zerwerfen ‘to separate’
ἅπτομαι ‘touch’
γαμοῦμαι ‘marry (of women)’45
συμβάλλομαι ‘contribute, help’
taka-st ‘succeed’
zāgtie-s ‘sneak (~go furtively)’
syne-s ‘seem’
volverse ‘to become’
finna-s ‘be on hand’

Once formed, lexemes often sustain further changes along the lexical-grammatical
continuum. This type of change is generally incremental, with gradual shifts toward the lexical
pole, from “less” to “more” lexical, as when additional morphological or semantic variation
modifies existing lexemes (Brinton and Traugott 2005, 96).
(3.7)

κόπτω ‘smite, strike, cut off’
κόπτομαι ‘strike or beat oneself [in mourning]’ > ‘mourn’ (intr.) > ‘mourn s.o.’ (tr.)

(3.8)

γράφω ‘write’
γράφομαι ‘take notes for one’s use, enroll oneself’ > ‘indict for public offense’

By their nature, lexemes are free to deviate from their original coinage. As a diachronic process,
this involves figurative extensions and other associative connotations that result in meanings that
are semantically idiosyncratic and lexically individuated. Thus, their synchronic outcome may no
longer reflect their original morphological contrast process, cf. Hellenistic κόπτω ‘cut off’ vs.
κόπτομαι ‘mourn s.o.’ in (3.7) and στέλλω ‘dispatch’ vs. στέλλομαι ‘avoid s.t.’ in (3.9)
(Geniušienė 1987, 149; Bauer 2003, 44; Nedjalkov 2007b, 319).
45

Also, τεκνοποιέω: Active (of women) ‘bear children’ vs. middle (of men) ‘beget children’.

63

(3.9)

στέλλω ‘make ready, send, dispatch’ > ‘gather, furl [as a nautical term]’
στέλλομαι ‘get ready, set out, journey’ > ‘keep away, avoid s.t.’

(3.10) ἀναστρέφω ‘overturn s.t., turn s.t. upside down’ > ‘turn back’
ἀναστρέφομαι ‘go up and down, roam’ > ‘stay, live’ > ‘conduct oneself well, behave’
The reason voice categories are prone to idiosyncrasy is closely tied to their semantic
function. In many cases, voice alternations (aside from the passive) entail substantial semantic
change to the meaning of the verb. Anticausative oppositions like those in (3.9)-(3.10) as well as
reflexives like (3.7)-(3.8) involve not only a change in event development (i.e. the semantic
nature of the action) but also a change in the number of participants. In Bybee (1985, 13), this
semantic impact is described in terms of relevancy. A category is considered relevant to the verb
if it directly alters its semantic content (Haspelmath and Müller-Bardey 2004, 1139-40).
For comparison, consider voice in contrast to tense. While voice is highly relevant to
lexical semantics, directly altering the nature of the action described by the verb, tense is less
relevant; it does not alter the meaning of the verb, but only distinguishes when it takes place. In
addition, a typical attribute of highly relevant categories like voice is their tendency to produce
more semantically distinct results. That is, they are more likely to create derived words that are
quite distinct in meaning from their bases. Thus, two forms of the same verb that differ in voice
are more semantically distinguished than two forms of the same verb that differ in tense. Over
time, semantically distinct oppositions are more likely to drift part as they take on further
semantic associations and unpredictable shifts in their lexical value. As they become lexicalized,
this often warrants the inclusion of separate senses and sub-senses in lexical entries to capture
their semantic shifts. Perusing any Greek lexicon illustrates this point in regard to voice. Note
that the same type of significant semantic change does not occur with distinctions in tense
(Bybee 1985, 20-23, 83; Haspelmath 1987, 13-14).

64

3.2 Grammaticalization of passive morphology
With lexicalization, language elements become more idiomatic as they move from less to more
lexical. With grammaticalization, elements move toward the grammatical pole as they develop
new grammatical functions, expanding their marking domain in the process (Brinton and
Traugott 2005, 99). Recall that in traditional conceptions of -(θ)η-, this particular voice form is
often considered to be uniquely passive. Table 1 in §1.1.2 illustrates the traditional divide in
voice morphology. For imperfective aspect, there are two morphological sets, one active, the
other non-active, or middle-passive. For perfective aspect, there are three sets. If the first is
active (-σα, -σας, -σεν) that leaves two non-active sets remaining. The second is sigmatic (first
aorist) morphology (-σαμην, -σω, -σατο); the third is -(θ)η- morphology (-(θ)ην, -(θ)ης, -(θ)η).
Traditional syntactic accounts impose the rules of a derived passive system onto sigmatic -σαand -(θ)η- forms, claiming that one (-σαμην, -σω, -σατο) represents middle voice with a reflexive
function and that the other (-(θ)η-) is evidence of a distinct passive voice with a uniquely passive
syntactic function. While this kind of isomorphism is appealing, it runs counter to the nature of
voice, from a typological perspective and in the diachronic development of -(θ)η- in Greek.
Two typological patterns in voice morphology are pertinent to -(θ)η-. First, the passive
function in IE and non-IE languages alike tends to arise as a secondary function from a variety of
source domains, e.g. statives, anticausatives, inactive auxiliaries,46 causatives, and generalized
subject constructions47 (Shibatani 1985; Haspelmath 1987, 35-42; 1990, 38-54; Givón 2002, 2346

Also called ‘periphrastic passives’ (especially in IE languages), these constructions combine a participial
form of the verb with an inactive auxiliary (or non-agentive verb), such as be, become, fall, happen (e.g. ‘getpassives’ like the window got broken). At first, such auxiliaries function as main verbs in their own right, but over
time they are re-analyzed as verbal auxiliaries or affixes (Haspelmath 1990, 38-9).
47
Generalized subject constructions are often referred to as ‘impersonals’; the subject pronoun typically has
a generic or indefinite referent like anyone, someone, man, or person (Haspelmath 1990, 49; Givón 2002, 24). Such
constructions are functionally similar to passives since the patient is topicalized and the agent is backgrounded.

65

25). As a result, the multi-functionality of passive forms reflects a diachronic chain of related
extensions. Over time, as new functional types are added, we see how two very different
functions (passive vs. reflexive) are included in the scope of a single form. This relates to the
second typological pattern. Voice morphology tends to be multifunctional (§2.2.1). From a
synchronic stance (a slice of language at one period), a marker of passive voice is more likely
than not to express additional functions, e.g. reflexive, potential passive, fientive, anticausative,
etc. (Shibatani 1985, 825-30; Haspelmath 1990, 37-49; Givón 2002, 207-14). Synchronic variety
is integrated with diachronic development (i.e. how a form originates and develops over time)
and is best considered in light of historical change. If diachronic development shapes the
synchronic system, then the functional variety in -(θ)η- and its diachronic history are two sides of
the same coin; they shed light on the same data (Sweetser 1990, 9).
The -(θ)η- form in Greek can be traced back to a stative suffix in PIE (-η- from *-eh1-), a
derivational morpheme denoting state predicates. Its integration into voice inflection is the result
of a gradual process, in which a lexical-derivational form, initially restricted to certain lexical
types, develops into a larger meaning class by way of lexical expansion. As it applies to more
and more lexemes, it loses specificity and expands its marking domain, culminating in a more
inflectional affix that expresses voice (middle-passive) and aspect (aorist perfective) in the verbal
system (Haspelmath 1987, 39-43; Jasanoff 2002/2003, 143-47; García Ramón 2014, 162).
Before considering the rise of -(θ)η- in Greek, it is useful to briefly examine its origins in
PIE. The earliest stages of classical IE languages reflect a voice contrast inherited from PIE,

66

active vs. middle-passive (Sihler 1995, 448). 48 There was no distinct passive form in PIE, nor
originally in its daughter languages (Szemerényi 1996, 283; Kulikov 2006, 62-3; Wackernagel
2009, 160).49 In Greek, -(θ)η- originated as a stative affix, a form used to derive stative verbs
from nominal-adjectival roots. The root *h1rudh ‘redness’ (Old Iranian rú ‘red’) provides an
underlying source for a number of suffixed derivations. In (3.11) stative *-eh1- represents one
part of a family of patterns (Jasanoff 2002/2003, 143-47, 165-67; Beekes 2011, 95-96, 252-56):
(3.11) Stative verb *h1rudh-éh1- ‘be red’ (inherited in Latin rubere, Old Iranian ruidid)
Adjective *h1rudh-ró- ‘red’ (passed down to Greek ἐρυθρός, Latin ruber)
Abstract noun *h1rudh-í- ‘redness’ (as in Latin rubi-dus ‘red’)
Since PIE nouns are declinable, they take different forms. One declension is instrumental
singular *-eh1 in *h1rudh-éh1 ‘with redness’, or as a non-verbal predicate: [x]…h1rudh-éh1 ‘[x]
is/was with redness’, ‘[x] is/was red’. This nominal form provides the basis for deriving stative
verbs: *[x]-eh1- ‘be characterized by [x]-ness’ or ‘be/become [x]’. Evidence of ‘e-statives’ is
manifest in several IE languages, e.g. Latin, Hittite, Greek, Old Church Slavonic, Lithuanian, etc.
The *-eh1- form traces a unique developmental path in each (Jasanoff 2002/2003).
3.2.1

-(θ)η- development

The adoption of stative -η- (*-eh1-)50 in Greek begins with its integration into two lexical types:
states and changes of state,51 both of which are intransitive (one-participant) event types, though

48

Middle-passive morphology (or, non-active) expresses both middle and passive functions. This includes
events that are volitional (e.g. reflexives) as well as those that are non-volitional (e.g. spontaneous processes:
burst/break or passives brought about by an external cause: be broken by).
49
If any IE languages developed a separate passive marker, they did so independently. Though see Kulikov
(2006, 62-81) for a discussion of a possible passive in early Sanskrit.
50
In its earliest stages, -θη- was just -η-, reflecting the *-eh1- origin. The fuller -θη- form is a proto-Greek
innovation with no direct PIE equivalent. Since there is no clear consensus on the origins of the full -θη- form, I
have refrained from including one here. Aorist -θη- and future -θησ- were later developments added to the original
aorist form: -η-. See Jasanoff (2002/2003), Hinge (2007), and García Ramón (2014) for discussion and reviews. For
a discussion of the future form (-θησ-) in Homer and Classical, see Allan (2003, 178-202).
51
In some cases, adoption into the Greek aorist stem may be the result of a back-formation from adjectival
derivatives. A verb like ἄγνυμι ‘break’ (IE root *ueh2ǵ) (Beekes 2010, 13) serves as an example. In early Greek,

67

further expansion to transitive (two-participant) verbs quickly follows. The earliest -η- aorists in
Greek appear to supplant older root aorists: τερση- ‘dry up’ (*ters- ‘dry’) corresponds to Vedic
Sanskrit root aorist tr̥ṣat; ἐμανη ‘became mad’ from aorist *mn̥-tó (Sanskrit áma-ta); ἐμίγη
‘mixed’ competes with older root aorist μίκτο (Jasanoff 2002/2003, 163; Beekes 2010, 1470;
García Ramón 2014, 159). Since diachronic change is generally a progressive process, this
replacement of older aorists with -η- aorists likely began among certain lexemes in a limited set
of types and then slowly expanded over time. The following discussion considers the semantic
conditions under which it spread, from Homeric to Classical and Hellenistic periods.
To note, the gradual expansion of -η- aorists coincides with a lengthening of its form, e.g.
productive -θη- in τράφθη ‘turned (intr.)’, ἐφιλήθη ‘be loved’ (Jasanoff 2002/2003, 161-67).
Because the fuller -θη- form was created early in Greek (attested in Homer) and used alongside
the -η- form to denote the same semantic event types, I use the -(θ)η- label.52
The first source for -(θ)η- is among intransitive change-of-state verbs. These are oneparticipant dynamic (non-static) events that entail a change from one state to another for the
subject participant.53 They interact with -(θ)η- in two ways, depending on verbal transitivity.

aorist participles like *ueh2ǵ-ént ‘broken’ were established as reinterpretations of old adjectives, formed with a
grammaticalized -ént (*-éh1-(e)nt-) in PIE (-ént was originally an independent nominal form but was later used to
form adjectives and participles *rudh-ént ‘red’; *kruh2-ént ‘bloody’). These participial forms were likely interpreted
as aorist perfective (rather than imperfective) because of the punctual (i.e. non-durative) semantics of the roots
(*ueh2ǵ ‘break’). The reinterpretation into verbal use (from adjectives) began its spread into other verbal stems. As
aorist participles with the form -é- (*-éh1-) were adopted among verbs, they replaced older root aorist participles
*ueh2ǵ-meno- ‘broken’. The finite -η- forms (-ην, -ης, -η) are possible back-formations of aorist participles (-ηντ-).
This gives a derivational schema: (a) adjective *ueh2ǵ-ént ‘broken’, (b) aorist participle *ueh2ǵ-ént ‘having become
broken’, (c) aorist finite stem *ueh2ǵ-é- (ἐάγη ‘broke’) (Jasanoff 2002/2003, 164-5; García Ramón 2014, 178-9).
52
The -η- and -θη- forms express the same semantic types; there is no evidence of a functional contrast.
The choice between the two was often determined by the formal shape of the verb and by analogy to similar lexemes
(García Ramón 2014, 152-153; Emde Boas et al. 2019, 177). See Allan (2003, 126-141) for a discussion of the
morphological features and semantic factors at play in their distribution in Homer and Classical Greek.
53
These include physical change, mental/psychological change, and change in location. García Ramón
(2014, 162) uses the term ‘telic-transformative’.

68

First, -(θ)η- aorists denote the punctual fulfillment of a change of state, with physical and mental
processes as well as body motion and collectives (García Ramón 2014, 163).54
(3.12) Change-of-state verbs
Physical process
Mental process
Body motion
Collective (of groups)

ἐξηράνθη ‘dried up’, ἐπάγη ‘got stuck’, ἐάγη ‘broke’
ἐμνήσθη ‘remembered’, ἐτάρπη ‘became full, enjoyed’
κλίνθη ‘bent aside, turned’, ἐάλη ‘crouched, cringed’
ἠγέρθη ‘gathered’, ἐμίγη ‘mingled, joined’

Verbs in (3.12) denote changes of state and the -(θ)η- form indicates that the change is
complete; the subject reaches the state expressed by the verb.55 Some are strictly intransitive:
ἀλήθη ‘wandered’ (ἀλάομαι ‘wander’); ἐκ...ῥύη ‘flowed out’ (ἐκ...ρέω ‘flow out’).56 Others
alternate with an active transitive counterpart, as in the causative/anticausatives pairs in (3.13).
(3.13) ἐχόλωσεν ‘made angry, provoked’ ἐχολώθη ‘became mad’
ἔτρεψεν ‘turned (tr.)’
τράφθη ‘turned (intr.)’
μειξα- ‘brought together (tr.)’
ἐμίγη ‘joined (intr.)’
This initial foray enables -(θ)η- to extend to passives.57 For both passives and changes of
state, the subject undergoes some kind of change expressed by the verb. The differences are that
passives entail an external force, but not necessarily a full change of state (πλήγη ‘be struck’;
δοθείη ‘be given’), whereas changes of state do involve a change of state but do not involve
external force (τράφη ‘grew up’). This extension of -(θ)η- morphology to passives was likely
bolstered by alternating causative/anticausative pairs: ἔαξε ‘break [x]’ ~ ἐάγη ‘[x] broke’ leads to
a passive reinterpretation ἐάγη ‘[x] was broken (by)’. Thus, -(θ)η- spread beyond events that
often, or necessarily, occur as internally-induced changes (σαπήηι ‘rotted’) to events that are
54

Physical processes involve inanimate entities that go through a physical change of state. Mental
processes, motion, and collectives involve animate participants that experience a cognitive change (mental process),
a change in location/posture (motion), and with collectives, crowds that disperse/converge. Labels used are from
Kemmer (1993) and Allan (2003). Allan adapts Kemmer’s labels to his study of the Classical Greek middle.
55
This is consistent with the aorist as a perfective aspect (as opposed to imperfective).
56
For a discussion of the construction ἐκ...ρέω see García Ramón 2014, 168-69.
57
This extension to passives began pre-Homer since -(θ)η- was readily used with two-participant passive
events in Homeric texts, e.g. ἐβλάφθησαν ‘was hindered (by)’; ἐτύπη ‘was hit (by)’.

69

often, or necessarily, brought about by external intervention (κρίνθη ‘be chosen’). In the passive,
the patient, as primary focus, is given subject status; the agent is often left implied as a defocused
participant, but if expressed occurs in a non-argument oblique phrase (3.14) or with non-subject
case marking (3.14) (Shibatani 1985, 831; Haspelmath 1990, 33; Jasanoff 2002/2003, 165;
García Ramón 2014, 169-71; Emde Boas et al. 2019, 455).
(3.14) (a)...δάμεν Τρώων ὑπὸ χερσίν ‘...were defeated by the hands of the Trojans’ Iliad 8.344
(b)...δάμεν Ἕκτορι δίῳ ‘...were defeated by godlike Hector’ Iliad 18.103
The second source for -(θ)η- aorists is states.58 Here, -(θ)η- functions as an ingressive, or
inceptive marker, expressing entry into a state, i.e. a state commences (Haspelmath 1987, 33).
With φαίν- (*bheh2 ‘shine’), the active/middle opposition reflects a caused vs. non-caused state:
φαίνω ‘make visible’ vs. φαίνομαι ‘be visible’ (Beekes 2010, 1545; García Ramón 2014, 160). In
the middle, imperfective φαίνομαι expresses a state, ‘the city of Priam is visible’ (Homer, Iliad
13.11). But when a state verb interacts with perfective (punctual) aspect, it produces an inceptive
reading, or onset of the state: ἐφάνη ‘became visible’ as in ‘then appeared a great portent’ (Iliad
2.307-9) (Comrie 1976, 19-20; Bybee 1985, 148). This is illustrated with μαίνομαι ‘is mad’ vs.
ἐμάνη ‘became mad’; χαίρω ‘is glad’ vs. ἐχάρη ‘became glad’ (García Ramón 2014, 172-78).59
Given the above, it is worth noting that -(θ)η- diachrony accords with wider typological
patterns. Its passive function rises secondarily from an originally intransitive source, specifically
change-of-state verbs (especially anticausative alternations) (Jasanoff 2002/2003, 164-65; Allan

58

States are non-dynamic or static predicates. They do not involve change, e.g. be, love/hate, be dry/wet
(Payne 2006, 344; Pavey 2010, 95).
59
The -(θ)η- forms ἐφάνη and ἐχάρη (from statives φαίνομαι ‘be visible’ and χαίρω ‘be glad’) were Greek
innovations and had no older corresponding aorist forms. Adopting the -(θ)η- aorist form allowed them to complete
a full verbal paradigm with imperfective and perfective forms (Ramón García 2014, 160, 172-78).

70

2003, 132-33).60 This highlights the central importance of the change-of-state/anticausative type,
not just as an inception point for -(θ)η- but its pivotal role in shaping -(θ)η- development.
From the original change-of-state/anticausative verbs, -(θ)η- spread in two directions: (1)
toward more agentive intransitives and (2) to a wider collection of passives. First, its extension to
other intransitives marks a slight departure from more a patient-like subject toward one that is
more likely to show a degree of volitionality (and animacy), especially among motion verbs.
Body motion involves a change in location or posture (ὁρμήθη ‘set out’) rather than an internal
physical change of state (ἐάγη ‘broke’). The subject is animate and tends to be volitionally
involved,61 though this is a scalar notion and varies with each lexeme.62
Second, -(θ)η- extended to passives. The subject is more patient-like; a non-volitional
participant undergoes an externally initiated event. A scalar notion applies since the passive may
or may not involve a complete change of state (δρύφθη ‘be torn apart’; ἑλκηθείς ‘be dragged
about’). The -(θ)η- form spread to a wider class of verbs, allowing it to span the continuum from
more derivational (lexically restricted) to more inflectional (lexically general). This means that
for state/change-of-state types the meaning of the verb is highly relevant and the -(θ)η- form is
semantically restricted to a particular lexical class. Its extension to the passive marks a move
toward a more lexically-general process. The passive is a syntactic realization and the meaning
of the verb is less relevant since it applies to any verbs that allow a two-participant conception.

60

Initially simply as -η- but its further expansion was likely reinforced by its lengthened form.
Body motion verbs involve a conflation of roles onto a single participant. The same entity that performs
the action (as agent) simultaneously undergoes the action (as patient). When someone sets out or bends over, the
subject is both the agentive source of the action as the one who volitionally performs the event. At the same time,
the subject is also the one who undergoes the changes that the action produces. This may partially account for the
spread from internal physical change to body motion. Change-of-state verbs are discussed in §4.3.
62
Agent-like ὁρμήθη ‘set out’ as in ‘they set out in pursuit’ (Iliad 10.359) vs. patient-like ἰδνώθη ‘bend
over, double up in pain’ as in ‘he doubled over and fell’ after being struck in battle (Iliad 13.618).
61

71

The scope of the -(θ)η- form includes a range of types, from those that are lexically
restricted (state/change-of-state) to those that are lexically general (passives), and from those that
are more patient-like (passives) to those that are more agent-like (body motion). Physical
change-of-state verbs occupy an intermediate space, serving as a semantic link between passives
and more agent-like changes of state (e.g. body motion).
Typologically, the change-of-state/anticausative type also plays a central role in the
development of passive and reflexive markers across a variety of languages (Shibatani 1985). As
noted in §2.2.1, a frequent pattern applies to anticausative-passive polysemy; the same form
marks both one-participant changes of state and two-participant passives (Haspelmath 1987, 28).
(3.15) Swahili

Anticausative vunj-ik- ‘break (intr.)’
Passive
it-ik ‘be called’

vunj- ‘break (tr.)’
it- ‘call’

The anticausative also connects passives with other more agent-like intransitives. In
Turkish, the same form (-Il-) marks passives, physical changes, and body motion (3.16) – much
like -(θ)η- in Greek (Geniušienė 1987, 327; Haspelmath 1987, 28; 2003, 225-6).
(3.16) Turkish Anticausative
Passive
Body motion
Greek Anticausative
Passive
Body motion

aç-il-mak ‘open (intr.)’
aç-mak ‘open (tr.)’
yap-il-ir ‘are made’
dik-il-mek ‘to stand oneself up’
ἐάγη ‘broke (intr.)’
ἔαξε ‘break (tr.)’
ἐβλάφθη ‘was hindered (by)’
ὡρμήθη ‘set out, rushed out’ ὥρμησα ‘set in motion’

There is also widespread correlation with reflexives, anticausatives, and passives; the
same form is used for all three functions. This is true for both IE (Latin, Spanish, Russian,
Romanian, Armenian) and non-IE languages (Hungarian, Arabic, Amharic) (Geniušienė 1987,
244-58, 308-20; Haspelmath 1987, 24-35; 1990).63

63

Also, anticausative-reflexive polysemy. One form marks both functions, but does not include the passive,
e.g. Quechua: anticausative paska-ri ‘open (intr.)’ ~ paska- ‘open (tr.)’; reflexive riku-ri ‘see oneself’ ~ riku- ‘see’.

72

(3.17) Arabic
Old Norse

Anticausative
Passive
Reflexive
Anticauative
Passive
Reflexive

ta-bayyanna ‘become clear’
ta-rabbā ‘be raised’
ta-labbasa ‘dress oneself’
hræð-sk ‘be frightened’
skeina-sk ‘get hurt’
klæð-sk ‘dress oneself’

bayyanna ‘make clear’
labbasa ‘dress (tr.)’
hræða ‘frighten’
klæða ‘dress’

The significance of the anticausative type lies in its semantic character. It is semantically
adjacent to other types, showing only minimal differences with each (Haspelmath 1987, 31). For
anticausatives and passives, the subject is a patient that undergoes a change; the difference is in
the energy source. For anticausatives, the energy is internal; for passives, it is external. The same
applies to anticausatives in relation to reflexives. Both represent one-participant events that are
internally induced rather than externally driven by a second more agentive participant. Likewise,
the subject in each undergoes some change – whether internal change or external manifestations,
such as a change from dressed to undressed (reflexive). The difference lies in how the subject is
involved in the event. With the anticausatives, the subject is more patient-like. But for reflexives,
the subject is more agentive, showing volitional instigation in bringing about the event.
To summarize: What begins as a PIE stative affix is adopted into Greek (-η- from *-eh1-)
as a lexical-derivational form, marking a limited spectrum within the lexicon. It is initially
integrated among change-of-state verbs, or one-participant (intransitive) events in which the
subject is a patient or undergoer, i.e. the participant that changes state. Through lexical expansion
(motivated by semantic similarities) the anticausative alternation plays a central role in -(θ)ηdevelopment, expanding to more agent-like events (body motion) and more patient-like ones
(passives). Or, to put it another way, it is characteristic of -(θ)η- aorists that the fulfillment of the
action leads to a change of state for the subject participant, whether this is among states
(φαίνομαι ‘be visible’), denoting entry into the state (ἐφάνη ‘became visible’), one-participant
changes of state (physical, mental, motion, collective) in which the subject reaches a state as a

73

consequence of a completed process (οἰνωθείς ‘got drunk’; ἐκορέσθη ‘be satiated, satisfied’), or
finally, among two-participant passives, wherein the subject undergoes a change as a result of
external force (ἔκταθεν ‘be slain’, σαώθη ‘be saved’) (García Ramón 2014, 151).
3.2.2

-(θ)η- expansion

A few observations may be made with regard to -(θ)η- development in Classical and Hellenistic.
Specifically, this concerns its semantic scope within middle voice and its attendant relationship
with older middle morphology as it expands in the Greek voice system. It was previously noted
that one of the patterns in -(θ)η- diachrony is its gradual displacement of older aorist forms. This
includes root aorists (βλῆτο supplanted by ἐβλήθην in Classical) as well as thematic aorists
(Homeric ἐπιθόμην gives way to Classical ἐπείσθην; Homeric ἐλιπόμην > Classical ἐλείφθην).
But it is the -σα- middle (-σαμην, -σω, -σατο) aorist that is of particular interest. Over the course
of Greek, from Homer to Hellenistic, -(θ)η- slowly expands its marking domain, largely at the
expense of the -σα- middle, such that by the Modern period, the -σα- form has fallen out of use,
having been fully displaced by -(θ)η- as the primary middle marker in the perfective paradigm
(Manney 2000; Allan 2003, 148; 169-73). The following discussion provides a brief outline,
tracing the contours of this expansion from Homer to Hellenistic.64
Recall that in Homer -(θ)η- marks a limited spectrum in the semantic middle domain.
Change-of-state verbs (ἰάνθη ‘became ripe, warm’) and passives (ἐβλάφθησαν ‘be hindered’) are
more patient-like middle types; the subject tends to be a (non-volitional) patient/undergoer. By
contrast, the -σα- middle is restricted to more agent-like middles; the subject is animate and tends
to be volitionally involved. The -σα- form is used for the following middle types in Homer:

64

For a more detailed account of this expansion in Homer and Classical Greek see Allan 2003, 126-77.

74

(3.18) More agent-like: -σα- middle form
Benefactive ἐδέξατο ‘received, accepted’
Grooming ἀλείψατο ‘anointed oneself’
Reciprocal μαχέσσατο ‘fought’

Speech act
μυθήσατο ‘spoke’65
Perception
γεύσασθαι ‘tasted’
Mental activity μητίσασθαι ‘devised’

More patient-like: -σα- middle form (change-of-state verbs)
Mental process χολώσατο ‘became angry’
Body motion ὡρμήσατο ‘set off, rushed out’
Collective
λέξασθαι ‘gathered’
Note that the -σα- middle is not used for physical proceses or passives; these are the
domain of -(θ)η- morphology. Nonetheless, the two forms do overlap. During any process of
change (e.g. a morphological shift from -σα- to -(θ)η-) there is necessarily a transitional period in
which either form is acceptable. Speakers do not adopt forms overnight and since the process is a
gradual one, it takes time for older morphology to fade out of use and new forms to become
predominant. In Homer, -(θ)η- usage extends into change-of-state verbs (mental processes and
body motion), thus producing newer variants alongside older forms (Allan 2003, 150-53):66
(3.19) -σα- and -(θ)η- variants among change-of-state verbs: Homer
Mental process
ἐμνήσατο ~ ἐμνήσθῆναι ‘remembered’
ἐχολώσατο ~ χολώθη ‘became angry’
ὀίσατο ~ ὠίσθη ‘thought’
Body motion
τανυσσάμενος ~ ἐτανύσθη ‘stretched out’
ὡρμήσατο ~ ὡρμήθη ‘set off’
τρεψάμενοι ~ τραφθῆναι ‘turned’
The shift into Classical and Hellenistic, brings with it two major consequences for
distributional patterns in voice morphology. First, -(θ)η- solidifies its position among more
patient-like middles in (3.20), taking the place of the -σα- form as the primary marking pattern
for change-of-state verbs (physical and mental processes, body motion, and collectives) along
65

Once again, labels used are from Kemmer (1993) and Allan (2003).
It should be noted that this kind of variation is not limited to -σα- and -(θ)η- alone. For a verb like ἐμιγη
‘mingle, join’, both -(θ)η- aorist μίγη and root aorist μίκτο are available variants: ‘he joined (μίγη) battle with the
Trojans’ (Iliad 5.143) and ‘he joined (μίκτο) his own throng’ (Iliad 11.354) (Allan 2003; 169-70; García Ramón
2014, 163-9). Later on, μίκτο falls out of use and -(θ)η- becomes standard, e.g. ἐμίγησαν ‘they mingled with the
nations’ (Psalm 105:35). See Allan (2003, 148-177) for further examples of morphological variation in the aorist.
66

75

with passives. This coincides with a gradual displacement of -σα- middle morphology among
these types, such that Homeric variants (ἐμνήσατο ~ ἐμνήσθῆναι ‘remembered’) see a shift to
more dominant -(θ)η- patterns (ἐμνήσθη) in Classical and Hellenistic periods.
(3.20) Patient-like middle types: -(θ)η- middle form is dominant
Body motion
ὡρμήθη ‘set off’; ἀπηλλάγη ‘leave, depart’; ἐπορεύθη ‘go, walk’
Collective
συνελέγη/συλλέχθη ‘gather’; συνεπορεύθη ‘go together, assemble’
Mental process
ἐφοβήθη ‘fear’; ἠγάσθη ‘admire’; ὠργίσθη ‘grow angry’
Physical process
ἐγενήθη ‘be born, come about’; πλήθη ‘become full’
Passive
ἐκλήθη ‘be called, summoned’; εὑρέθη ‘be found, caught’
This results in a formal split in the perfective aorist middle paradigm; -(θ)η- morphology
is dominant among more patient-like middles, whereas -σα- morphology is dominant for more
agent-like middles. Examples (3.20) and (3.21) reflect broader distributional patterns for -σα- and
-(θ)η- morphology for Classical and Hellenistic periods (Allan 2003, 154-56).
(3.21) Agent-like middle types: -σα- middle form is dominant
Perception
ἐθεάσατο ‘look at’; ἐπεσκέψατο ‘look at, inspect’
Speech act
ἐψεύσατο ‘lie, speak false’; προσεύξατο ‘pray’
Mental activity ἐβουλεύσατο ‘deliberate, decide’; ἡγήσατο ‘hold that, regard’
Reciprocal
ἠγωνίσατο ‘struggle, fight’; ἐμαχέσατο ‘fight’
Grooming
ἐλούσατο ‘wash’; ὑπεδήσατο ‘put on shoes, bind on feet’
Benefactive
ἠργάσατο ‘work, work at’; ἐκτήσατο ‘acquire’
The -σα- and -(θ)η- forms divide up the semantic space of the middle, as in (3.20)-(3.21).
But the boundary between them is not hard and fast. In fact, -(θ)η- shows sporadic extension into
more agent-like middles.67 The -σα- form is dominant, with -(θ)η- encroaching on a few lexemes.
(3.22) Agent-like middle types: -(θ)η- variants
Perception
ὠσφρήσατο ~ ὠσφράνθη ‘smelled’, ἔδρακον ~ ἐδέρχθη ‘saw’
Speech act
ἐδεήθη ‘ask, beg’, ἐμνήσθη ‘mentioned’
ἐμβριμησάμενος ~ ἐνεβριμήθη ‘warned sternly, scolded, rebuked’
ἠρνήσαντο ~ ἠρνήθη ‘denied, refused, declined’
67

The term ‘sporadic extension’ is used to suggest that -(θ)η- does not show widespread distribution among
these types. In fact, only a few lexemes in each type show morphological variation between an older aorist form and
a new -(θ)η- variant. The majority of lexemes in these more-agent like middles receive -σα- middle morphology in
Classical and Hellenistic, with a limited distribution of -(θ)η- forms.

76

ἀμείψατο ~ ἠμείφθη ‘answered’, ἀπεκρίνατο ~ ἀπεκρίθη ‘answered’
ἐλοιδορήσαντο ~ λοιδορηθείς ‘railed at, reproached’
κριθῆναι ‘quarreled, disputed/pleaded in court’
ἀπελογήσατο ~ ἀπελογήθη ‘spoke in defense, pleaded’
Mental activity ἐπιλεξάμενος ~ ἐπιλεχθείς ‘thought about’
ἐπελογίσαντο ~ ἐπιλογισθέντες ‘considered’
διεκρίθη ‘differentiate, judged against, opposed one another’68
ἐμηχανήσατο ~ μηχανηθείσης ‘contrived, devised’69
Reciprocal
διελέξατο ~ διελέχθη ‘conversed with, discussed, argued’
ἐμαχέσατο ~ μαχεσθῆναι ‘fought’70
ἁμιλλήσαιντο ~ ἁμιλληθείς ‘vied for, contended with’
συνειστιάθησαν ‘dined with, feasted with’
Instances in (3.22)-(3.23) illustrate -(θ)η- variants that arise in Classical and Hellenistic
usage for perceptions, speech acts, mental activities, and reciprocals, along with a few grooming
and benefactive middles (3.23) that newly arise in Hellenistic.
(3.23) Grooming

Benefactive

ἐβαπτίσατο ~ ἐβαπτίσθη ‘dipped, washed oneself’
ἐκτιναξάμενος ~ ἐκτιναχθήσομαι ‘shook s.t. off oneself’
ἡγνίσαντο ~ ἁγνίσθητε ‘cleansed, consecrated oneself’
θερμανθεὶς ‘warmed oneself, got warm’
ὑποκρίνασθαι ~ ὑποκριθῆναι ‘pretended, acted hypocritically’

Middle morphology, in perfective and imperfective paradigms, represents a semantic
continuum rather than a dichotomy. In this continuum, middle morphology includes both middle
and passive semantics in its scope. Though, it is notable that the rise of the -(θ)η- form and its
integration into voice morphology has significant consequences for the synchronic structure of
the Greek voice system. For the imperfective paradigm (relevant to present, imperfect, perfect,
and pluperfect forms), middle marking is consistent across all middle types. The same form
68

The middle form of διακρίνω has two primary senses in Hellenistic; both involve opposition or conflict
based on differing judgments. One is a volitional mental activity ‘dispute with or differentiate between one another’,
as in ‘Have you not made distinctions (διεκρίθητε) among yourselves?’ (James 2:4). The other is a non-volitional
mental process. Conflict results from inward dispute (‘doubt, hesitate, be at odds with oneself’), as in ‘No distrust
made him waver (διεκρίθη) concerning the promise of God’ (Romans 4:20).
69
‘She saw that her endeavors would not succeed…’ μὴ ἀποσκευῆς αὐτῇ μηχανηθείσης τῶν γνησίων τοῦ
Φραάτου παίδων ‘unless she could contrive how to remove Phraates’s legitimate sons’ (Josephus Antiquities 18.41).
70
The -σα- form remains the primary pattern in Hellenistic, though -(θ)η- continues to expand, e.g. ἐγὼ
βούλομαι μαχεσθῆναι ‘I want to fight’ (Josephus Antiquities 6.185).

77

expresses agent and patient-like functions of the middle. But in the perfective paradigm
(aorist/future), middle marking diverges from this pattern. The dual means of middle marking,
via -σα- and -(θ)η- forms, results in an overlapping of middle types. Each form approaches
middle semantics from opposing ends of the continuum (agent-like middles for the -σα- form,
patient-like middles for the -(θ)η- form) as they meet and overlap in the central space.71

71

See Emde Boas et al. (2019, 464) for a chart illustrating these usage patterns.

78

4. Event construal and the nature of voice
Chapter 1 lays out a descriptive account of Greek middle-passive syntax and semantics relative
to event conceptualization. Prefaced by a brief summary of event structure and construal, the
grammatical description begins with the patient-oriented middle-passive uses in their contrast
with the prototypical transitive event and moves progressively along the transitivity continuum,
ending with reciprocal, grooming, and benefactive events.
Before examining each middle type in turn, it is worth discussing certain parallels from
embodied experience that address the connections between how we conceive events and how we
express them in language. Grammatical voice distinctions find their conceptual basis in how we
understand and categorize events and the participants involved in them.
A few conceptual notions may furnish a foundation for talking about voice distinctions.
Some of our most fundamental cognitive patterns arise from embodied experience of movement
and interactions in space. These conceptual schemas are cognitive processing patterns that help
us to engage in the world by taking vastly different experiences we have in the course of a day
and understanding them through recurring elements. These patterns and processes help us to
comprehend two basic themes in events, participants and relationships. In relationships, two
schemas are pertinent: one and two-participant interactions (Langacker 2006, 116-7). Some
relationships involve one focused participant, which may occupy a location (κάθημαι ‘sit’) or
change that location (πνίγομαι ‘drown’, ἐβυθίσθην ‘sink’). A single participant may exhibit a
stable property (ξηρός ‘dry’) or undergo a change in property (ξηραίνομαι ‘dry up’). A person can
experience a state (κοίμησις ‘sleep’) or a change in state (διαγείρομαι ‘wake up’).
Figure 1 illustrates one-participant schemas. Such relations may be static, in which a
single participant is in a static location, property, or experience, as in the left column.

79

Conversely, they may also be dynamic; the participant goes through a dynamic process with a
change in location, property, or experience, as on the right.72

Figure 1

Types of event conceptualization

Either type (static/dynamic) can be conceptualized without involvement from an outside
force. Alternatively, a dynamic version of these prototypes may be conceptualized as a more
elaborate two-participant relation, representing common types of causation. Arrows in Figure 2
indicate exertion from an external force that induces a change of state in the second participant.

Figure 2

Two-participant caused change (Langacker 2006, 117)

Relational schemas presuppose that participants relate together in a given event. Because
of this, we can talk about participant roles. Participant roles have to do with how participants are

72

Figures in chapter 4 are adapted from Langacker (2006, 115–37). In figure 1, circles represent locations;
squares are properties; solid lines are physical changes; and dashed lines are mental/psychological changes.

80

involved in an event.73 The more active participant, the agent in Figure 2, is the energy source,
the one who induces a change in other participants. Opposite to the source is the endpoint, the
participant changed by the action. The patient in (a) undergoes a change of state; the mover in (b)
changes location. The experiencer in (c) goes through a cognitive or emotional change of state.
Conceptual schemas shape our thinking and avail themselves of expression in language.
They are readily accessible as the prototypes for fundamental grammatical categories. As such,
conceptual schemas mediate between situations in the world and grammatical expressions in
language. Our body-based understanding of physical objects supplies the prototype for the
linguistic category of nouns. And our body-based understanding of causation/physical force
supplies the prototype for the linguistic category of active transitive events.
4.1 Semantic transitivity
In a basic active transitive clause in (4.1), the agent is the starting point or energy source and the
patient is the endpoint or energy goal (Kemmer 1993, 50-51; Næss 2007, 28).
(4.1)

οἱ στρατιῶται τοῦ πρώτου κατέαξαν τὰ σκέλη
The soldiers broke the legs of the first man (John 19:32)

This transitive event type provides the basis for most two-participant events in language. It arises
through our embodied experience of physical force and energy transfer, where humans, as agents
in the world, act on other entities by physical contact. A volitional participant (agent) causes a
change of state in another participant (patient). This is the prototypical transitive event.74

73

Participant roles (also termed thematic or semantic) are rooted in semantic and pragmatic relations
expressed in language, regarding how participants relate to one another in a given event. Participants roles differ
according to the type of action. As such, roles like experiencer, stimulus, recipient, beneficiary, agent, and patient
are useful in recognizing different types of relationships among participants in voice distinctions (Kemmer 1993, 8).
74
It is best understood as an experiential gestalt because it involves a cluster of properties that we
understand as a patterned whole (Lakoff 1987, 54–55).

81

As a prototype category, the basic two-participant relation provides a framework for
organizing knowledge about the world and understanding different event types in our shared
human experience. Because of this, it is concerned with the construal process, or how we think
about actions. Because the construal process is dynamic, we can think about events from a
variety of perspectives and express them in language accordingly. Given the default active in
(4.1), we may construe the event in different ways. One strategy focuses on the man with the
broken legs by expressing the same action in a passive clause: ‘The legs of the first man were
broken by the soldiers.’ Another strategy focuses on how the soldiers work in a group to carry
out the action: ‘The soldiers together broke the legs of the first man.’ Each expression shapes
how we see, or process, the event. In the grammatical resources of a given language, the same
situation can be coded (i.e. formally expressed) in multiple ways with alternate grammatical
devices that reflect how a speaker conceives an action (Næss 2007, 11- 17).
Each grammatical structure has conceptual content and imposes imagery on an event.
Coding and construal are interdependent processes (Langacker 1987, 294). A situation’s
construal determines whether a linguistic structure may code it. In turn, linguistic structures
embody conventional imagery and thus impose a certain construal on the situation they encode.
This construal process is inevitably present in all language use, reflecting the fact that languages
provide various means (via grammatical structures) for categorizing events and participants in
different ways. Speaking entails choice; every utterance construes an event in some manner,
imposing an interpretation on the content involved. As soon as people begin to speak, they are
participating in a meaningful construal process; each linguistic utterance is an interpretive act.

82

4.2

Adjustments in energy and attention

The transitivity continuum provides a framework for understanding voice. Voice alternations
offer the grammatical means for expressing different conceptualizations of events. In this, voice
is a communicative tool, one among many in language, for interpreting events in different ways.
Specifically, voice involves two principal motivations in the construal process. One is energy
flow in the event; the other is the focus of attention we give to different aspects of an event.
First, every action involves some type of energy transfer. For this to happen, there must
be some type of change. Bare states like ‘I live here’ lack change and thus also energy transfer.
For the prototypical transitive event, there are two participants: source and endpoint. Imagine a
girl kicking a ball. The girl, as energy source, makes contact with the ball, as endpoint. Energy is
transferred from her foot to the ball. The ball (patient) receives the energy and undergoes a
change in location as a result; it flies through the air from one point to another farther away. In
this conception, no change in location is profiled for the prototypical agent (energy source); the
agent is unaffected by the event. As the endpoint, the prototypical patient receives the energy and
is altered by it. We can follow the energy transfer from agent to patient and observe its effects.
This basic event type is typically coded in the prototypical active transitive clause so that
we mentally scan it from cause to effect, from source (coded as subject) to endpoint (object).
This pattern reflects the nature of our body-based experience with causation: In cause-effect
relationships, someone does something that causes something else to happen; we see the
consequences of this action take place in the world (Langacker 2006, 123).
Two points regarding the prototypical active transitive are important to keep in mind visà-vis voice alternations. The basic alignment involves two distinct participants playing two
distinct roles (Kemmer 1993, 65-66). Distinctness of participants involves a conceptual

83

separation from both (a) the background and (b) one another in terms of the roles in the event.
The prototypical transitive involves two distinct participants; both are salient in event conception
and distinct from the general background.75 The subject, as energy source, plays a volitional
agent who causes the event to take place and does not experience a change of state in it. The
object, as endpoint, plays a non-volitional patient who receives the energy transfer and
undergoes a change of state. Figure 3 depicts the active transitive prototype (Næss 2007, 23, 30).

Figure 3

Prototypical transitive event

The second motivation in voice involves a visual metaphor highlighting the scope of
attention. All languages provide structural schemas for one- and two-participant clauses. Each
schema highlights one participant as primary focal point. In an agent-oriented system like Greek,
primary focus is conferred on the most agent-like participant as the source of energy (Langacker
2006, 121-23). When the template is metaphorically extended to non-agentive interactions, the
more agent-like figure (e.g. experiencer in mental events) is naturally chosen as primary focal
point and energy source, with the more patient-like participant as a secondary focal point.
As we saw in (4.1) with the soldiers breaking the man’s legs, we can shift attention to
different aspects of an event. In a two-participant clause, such as the active transitive prototype,
there is generally a primary and secondary focal point (Langacker 2006, 119).
(4.2)

ἐμέθυσεν αὐτóν
He [David] made him [Uriah] drunk (LXX 2 Samuel 11:13)

75

This salience of participants will become evident in §4.2.1 with the passive alternation.

84

(4.3)

ἐμεθύσθη Οὐρείας
Uriah became drunk
An expression like (4.2) selects a two-participant relation as locus of attention. Two on-

stage participants play two distinct roles. David, as primary figure, is the energy source or
starting point, for the action. Uriah, as secondary figure, plays the endpoint role. When the action
begins, that is, when the agent does something, the spotlight, or primary focal point, is on the
energy source. But since the process highlights the agent’s activity in regard to its interactions
with a second figure, we follow the arc of the action as its effects play out on a distinct patient.
The focus of our attention, or the spotlight, shifts from primary figure to secondary participant as
we see the consequences of the action unfold.
But if a speaker chooses to narrow the focus from two participants to one, then there is
just one on-stage figure in the spotlight. Removing the role of the agent in (4.3) shifts the
primary focal point from agent to patient and puts in profile the patient’s change of state from
sober to drunk. This gives the event a new starting point for mentally accessing it and allows a
speaker to highlight the change undergone by the primary figure. The starting point and endpoint
of the action are conflated onto a single participant. The natural consequence of this is that the
patient participant is now the primary focal point as the most salient for the event and has
undivided attention. In contrast to a two-participant event, we can already see how removing the
agent and thus changing the number of participants shifts the centre of attention for an event.
The transitive prototype (a) profiles a two-participant relation with an agent-patient
interaction and (b) highlights the most agent-like figure as the primary focal point, or energy
source, for the event, with the patient as a secondary figure, or energy endpoint (Langacker 2006,
121-29; Shibatani 2006, 221-222). Departures from the transitive prototype regarding the

85

interplay of these two parameters give rise to voice alternations. Voice involves two kinds of
motivation for how we construe events in different ways:
• Adjustments in the energy source and energy endpoint for the event76
• Adjustments in the scope of attention
o The natural consequence of shifting the focus of attention in an event is a change
in the relative salience of participants (Langacker 2006, 121; Næss 2007, 23).
Situations that deviate from the transitive prototype with respect to these two parameters
are often encoded by marked voice constructions to explicitly signal their conceptual difference.
Active voice supplies the basic voice type, while middle marking indicates that an event departs
from the basic transitive prototype in some way (Langacker 1987, 125; Shibatani 2006, 217-69;
Næss 2007, 17). Each of these voice parameters play out in middle event types discussed below.
4.2.1

Alternations in voice: Shift in primary focus

The active transitive clause represents the most basic alignment for two-participant interactions.
The energy source (agent) for the event plays the role of default primary focal point, with the
energy endpoint (patient) as a secondary focal point. Given this type of default configuration,
this creates a need for an alternate construal, one in which discourse coherence favors focusing
on the secondary figure instead. Passive structures allow for grammatical flexibility in language
(reflecting our cognitive construal abilities), for expressing two-participant relations as endpointoriented processes rather than source-oriented events (Langacker 2006, 125; Sansò 2006, 238).
Selecting the endpoint patient as primary focal point increases its relative salience for the
interaction, thereby highlighting the core process that the patient undergoes. The passive in (4.4),
coded with the -(θ)η- form, represents the most patient-like event on a continuum from agent to

76

Terminology employed in cognitive linguistics is generally trajector (TR) and landmark (LM). The TR is
the starting point or initiating figure for the event. The LM is the goal/endpoint, the one who receives the energy
transfer and is changed by it (Langacker 1987, 304–29). In a basic active clause, the subject is the TR as the primary,
most salient figure. The object as LM is secondary figure, or second most prominent participant.

86

patient-like event types. The primary figure is the participant who undergoes a change (brought
about by external force) and is not volitionally involved in the event (Langacker 2006, 127).
(4.4)

Ἡρῴδης ἰδὼν ὅτι ἐνεπαίχθη ὑπὸ τῶν μάγων
Herod saw that he was tricked by the wise men (Matthew 2:16)
The conceptual content evoked in the passive profiles the same overall interaction as the

active transitive (The wise men tricked Herod). But in the passive, the agent is less relevant or
salient for event construal. The role of the wise men in tricking Herod is downplayed in the scene
so that Herod’s anger and subsequent action becomes topicalized as focal point. Attention is
drawn to the patient (coded as subject), indicating that the relevance of the agent (coded in an
oblique phrase) is comparatively low, and becomes blurred in the background (Shibatani 2006,
248). The agent is still the initiating entity (source) and the patient remains the one who receives
that energy (endpoint). The key difference for the passive is a shift in the spotlight from agent to
patient so that the agent becomes a non-distinct participant, blurred in the background, and no
longer salient. Selecting a particular element as focus of attention in a profiled relationship
naturally enhances the aspects of the interaction that it underlies. If the active highlights the
agent’s activity in its interaction with a patient, then the passive highlights the central process
that the patient undergoes. In (4.4), it is the fact that Herod was tricked that makes him angry and
his subsequent response to that anger is highly salient for event construal: He decides to kill the
children of Bethlehem (Langacker 2006, 217).
The passive exemplifies but one alternative construal based on a shift in attention from
source to endpoint. Another conceptual alternation is achieved by changing energy flow. This
choice in energy source and endpoint, i.e. choosing which participant plays the starting point or
endpoint for the energy transfer, also diverges from the active transitive prototype. In the

87

perfective paradigm, the -σα- and -(θ)η- forms in (4.5) represent two kinds of deviation from the
basic active pattern, one for energy transfer (b) and one for focus of attention (c).
(4.5)

(a) Aorist active
ἐφύλαξεν ‘guard s.o./s.t.’
(b) Aorist -σα- middle ἐφυλάξατο ‘keep for oneself’
(c) Aorist -(θ)η- middle ἐφυλάχθη ‘be guarded (by)’

Agent-patient interaction
Change in energy endpoint
Shift in primary focus

The active in (a) profiles an event that ends with a distinct endpoint recipient. The -σαmiddle in (b) profiles an event that alters the energy endpoint (discussed more fully in §4.5.3).
The action takes place with the source as the intended endpoint. The spotlight is drawn to the
primary figure as source and endpoint/recipient of the action. The -(θ)η- middle form in (c) shifts
the focus of attention; the spotlight is on the patient rather than the agent’s activity.
The active transitive prototype forms the basis for understanding energy flow from source
to endpoint. The relational parameter of affectedness is thus implicated in the scale of transitivity
(Beavers 2011, 362). The more specific an event is about a participant’s progress along a scale of
change, the higher the degree of affectedness for that participant. Events that express a full
change of state (κλάω ‘break into parts’) entail a higher degree of affectedness than those that
involve only surface contact or impingement (παίω ‘hit’). This difference in scalar change is
illustrated with active-passive alternations in (4.6)-(4.7).
(4.6)

πειράζω ‘test [x]’
ἀπόλλυμι ‘destroy [x]’

πειράζομαι ‘[x] is tested (by)’
ἀπόλλυμαι ‘[x] is destroyed (by)’

(4.7)

βλέπω ‘see [x]’
ἀκούω ‘hear [x]’
εὑρίσκω ‘find [x]’

βλέπομαι ‘[x] is seen (by)’
ἀκούεται ‘[x] is heard (by)’
εὑρίσκονται ‘[x] is found (by)’

Highly transitive events (i.e. those closer to the semantic transitive prototype) in (4.6)
entail a specific result state for the energy flow of the event from source to endpoint. But lower
transitivity events in (4.7) involve no necessary change for the endpoint participant. This is

88

especially true for perception verbs (βλέπω ‘see’, ἀκούω ‘hear’). Rather than expressing an
agent-patient interaction, verbs in (4.7) involve an experiencer-stimulus relationship (cf. §4.4.3).
In such cases, an experiencer, as primary focus, directs his/her attention toward a
stimulus. The stimulus, as secondary figure, is the entity that prompts attention or sensory input
for the experiencer. But unlike a prototypical patient, the stimulus remains wholly unaffected by
the perceptual interaction. Thus, with a passive alternation (βλέπομαι ‘be seen’, ἀκούεται ‘be
heard’), the primary focus for the event shifts from the experiencer to the stimulus, a wholly
unaffected entity as the most salient for event conception.
What this suggests is that subject affectedness, as a gradable and relational parameter,
does not apply in the same way to every passive clause. Passives with verbs like ἀπόλλυμαι ‘be
destroyed’ implicate a change of state for the subject and thus a high degree of affectedness. But
passives with verbs like βλέπομαι ‘be seen’ entail no such change for the subject and thus express
little to no affectedness for the (stimulus) subject participant.
Despite differences in affectedness, passive construals in (4.6)-(4.7) do share two things
in common: (1) They involve two-participant relations that imply the conceptual presence of an
external source of energy, even if that energy source is downplayed in event conception. And (2),
the endpoint participant is brought into the spotlight as the onstage focus. Thus, passives
represent a change in attentional focus from source to endpoint. In the discussion that follows,
the rest of the middle event types are broken up into smaller sets, those that are more patient-like
and thus closer to the traditional passive type (§4.3), and those that are more agent-like and thus
closer to the active transitive prototype (§4.4- 4.5).

89

4.3

Alternations in voice: External vs. internal energy

Within the semantic space of the middle-marking domain, imperfective middle morphology in
Greek is formally consistent across all middle event types. The same form is used for reflexive
semantics as well as the passive function. The perfective paradigm, however, maintains two
middle forms in the Hellenistic period.77 In early Greek, the -(θ)η- form was restricted to
anticausatives (physical, mental, body motion, collectives) and passives. These two types occupy
the more patient-like end of the spectrum. But during Classical and Hellenistic periods, the -(θ)ηform began to expand its scope, so that -(θ)η- sporadically extended to other middle types, i.e.
those that are primarily expressed by the -σα- middle: reflexive, reciprocals, etc. More agent-like
events are discussed in §4.4-§4.5. For now, events in §4.3 share a property with the passive type:
Anticausative alternations include a marked choice in the focus of attention that departs from the
prototypical active. But, unlike the passive type, these middle types also alter energy flow for the
event, changing from a two-participant caused relation to a one-participant change of state.
Change of state: Physical process78

4.3.1

Recall from Figure 3 that an external cause brings about a change to a second participant. It may
be a change of state, experience, or location/posture. In each case, primary focus is on the agent,
as source and starting point for the event, causing a change of state in the patient, as secondary
focus and energy endpoint. Active voice is the default expression for such caused events.
Active clauses in (4.8) and (4.9) represent the default formal expression used for events
in which an agent (source) performs an action that induces a change of state in a distinct patient
77

I focus on the -(θ)η- and -σα- middle aorist forms here, but there are of course thematic and root aorist
forms as well. See Emde Boas et al. (2019, 464) for a verbal chart illustrating the use of such forms.
78
The labels in this and the following sections in ch. 4 are taken from Kemmer (1993) and Allan (2003):
Physical process, body motion, collective motion, mental process, mental activity, speech act, perception, reciprocal,
grooming, and self-benefactive.

90

(endpoint). The syntactic structure in each reflects a two-participant caused relation, with the
energy source encoded as subject and energy endpoint as object.
(4.8)

Active caused change
Λάβε καὶ κατάφαγε αὐτό [τὸ βιβλαρίδιον], καὶ πικρανεῖ σου τὴν κοιλίαν
Take and eat it [the scroll], and it will make your stomach bitter (Rev. 10:9)

(4.9)

καὶ τῶν κολλυβιστῶν ἐξέχεεν τὰ κέρματα
He also poured out the coins of the moneychangers (John 2:15)

In (4.10) and (4.11) a change in voice morphology alters event conception.
(4.10) Middle spontaneous change
ἐπικράνθη ἡ κοιλία μου
My stomach grew bitter (Rev. 10:10)
(4.11) ῥήγνυνται οἱ ἀσκοί, καὶ ὁ οἶνος ἐκχεῖται
the skins burst and the wine spills out (Matthew 9:17)
Middle voice expresses events in which a single focused participant undergoes a change
of state. The patient (object) of the causative active becomes the subject of the anticausative
middle. Removing the role of the causal agent shifts the focus of attention. The participant that
undergoes a change of state becomes primary focus and starting point for accessing the event.
This anticausative middle event type is a valency reducing process. An active (two-participant)
conception with an agent and patient is reduced to a middle (one-participant) construal with a
single patient participant undergoing a change of state.
Narrowing the scope of attention to one participant heightens its relative salience in event
conception. Syntactic structures in (4.10)-(4.11) signal that the patient is the primary focal point
and starting point for the action.79 Each is expressed as an intransitive clause with a single
participant. The subject plays the role of patient, as the one who undergoes a change. Energy

79

Alternatively, the causative active widens the scope of attention to include not just a changed participant
but an external causer that brings about that change, as agent (Haspelmath and Müller-Bardey 2004, 1141).

91

source and endpoint roles are conflated onto a single participant – the energy for the event is not
an external force but is internal to the changed participant.
This is why the physical process type cross-linguistically includes events usually
conceived as lacking an identifiable external force: grow, rust, rot, and other natural
physiological processes. But keep in mind that physical processes may also be formulated using
active morphology, as in (4.8), to express a causative version of the same state of affairs. Almost
all physical processes in Greek have a causative counterpart. Examples of the causativeanticausative contrast and middle-only physical processes are in (4.12).
(4.12) Active/Causative
ἀλλοιόω
make different, alter
ἀπόλλυμι
destroy
ἀφανίζω
make disappear
βυθίζω
cause to sink
(δι)εγείρω waken, rouse s.o.
δυναμόω
strengthen
δύω
cause to sink, plunge in
ἐκχέω
cause to pour out
(κατα)καίω burn, consume (tr.)
κρεμάννυμι cause to hang, hang up81
πνίγω
strangle, choke (tr.)
ῥήγνυμι
break (tr.)
σήπω
make rotten
σκοτίζω
make dark
τήκω
melt (tr.)
τρέφω
cause to grow, nourish
φθείρω
destroy
φλογίζω
ignite
φύω
cause to grow, beget
ψύχω
make cold, dry, refreshed

80
81

Middle/Anticausative
ἀλλοιούμαι
change (intr.)
ἀπόλλυμαι
perish, die
ἀφανίζομαι
disappear
βυθίζομαι
sink (intr.)
(δι)εγείρομαι awaken
δύναμαι
be/become strong > be able80
δύομαι
sink, set (of the sun)
ἐκχέομαι
spill out
(κατα)καίομαι burn (intr.)
κρέμαμαι
hang down, hang from
πνίγομαι
drown, choke (intr.)
ῥήγνυμαι
burst, break (intr.)
σήπομαι
rot
σκοτίζομαι
become dark
τήκομαι
melt (intr.)
τρέφομαι
grow
φθείρομαι
perish
φλογίζομαι
burn
φύομαι
grow
ψύχομαι
become cold, dry, refreshed

The middle expression is the older form; the causative active is a later Hellenistic formation.
In LXX, also κρεμἀζω ‘cause to hang, hang up.’

92

Media tantum – Physical process
γίνομαι
come to exist, be formed82 διαγίνομαι
κατιόομαι become rusted
ὀπτάνομαι

pass, elapse (of time)
appear, become visible

Μany media tantum (middle-only) verbs have semantics that refer to physical processes
that do not readily include an alternative causal expression, as with events that, from a folk
perspective, are less likely to have a ready external explanation, or the origin of their existence is
socially expected to lack an external cause, as with passing of time or coming into existence.
In contrast to physical processes, the passive construal implies the presence of an external
cause that brings about the event. The agent is conceptually present, even if it is not syntactically
expressed. Although the passive type often involves events that come about by external cause, no
clear demarcation is evident. At times, it may be difficult to tell if an external source is implied
or if it is construed as a one-participant event brought about spontaneously, as in (4.13).
(4.13) καὶ ὁ οὐρανὸς ἀπεχωρίσθη ὡς βιβλίον ἑλισσόμενον
And the sky split apart like a rolled-up scroll (Rev. 6:14)
Deciding to interpret (4.13) as a spontaneous physical process vs. a passive relies on the
prominence the reader gives to the supernatural context of the Revelation narrative.83 Natural
processes that “just happen,” earthquakes, floods, etc., are taken as spontaneously occurring
events, but a supernatural narrative like Revelation opens up the possibility for a passive reading.
In contrast to the active transitive, the anticausative middle narrows the scope of attention
so that the onstage focus is on the participant who undergoes a change of state. This shift in
attention heightens its salience so that it becomes the primary focal point. As the central onstage
figure, the patient undergoes an internal process of change that occurs without external force. For
82

Many change-of-state verbs have active perfects, expressing the state that results from the process of
change: γίνομαι (perfect γέγονα); ἀπόλλυμαι (perfect ἀπόλωλα); σήπομαι (perfect σέσηπα); φαίνω (perfect πέφηνα).
83
Some physical process verbs denote events that may be brought about spontaneously or by external
force, e.g. πνίγομαι ‘drown’ (spontaneous) or ‘be suffocated by’ (passive); φλογίζομαι ‘burn up’ (spontaneous) or ‘be
set on fire by’, ‘be burnt by’ (passive); φθείρομαι ‘perish’ (spontaneous) or ‘be destroyed (by)’ (passive).

93

the passive type, the patient is also the central figure, but the role of the external force (while
downplayed) is still conceptually present. For both types, the focus of attention remains on the
core process that the patient undergoes, thus highlighting similar facets of event development.
For the anticausative type, the conceptual content (or kind of event) is altered. Rather
than an agent causing a change in a patient, the anticausative removes the agent and presents the
event as occurring without external force. The starting point for the action is the patient. Because
attention is not divided between two distinct on-stage participants, the primary focus naturally
falls on the remaining participant. As both source and endpoint, energy flow is construed as
internal to the process. The anticausative type can be contrasted to the active transitive by means
of an image schema. In Figure 4, a two-participant active is shown in contrast to the oneparticipant event expressed via middle morphology.

Figure 4

Caused change vs. spontaneous change

Active expressions in (4.8)-(4.9) correspond to the figure on the left; an agent (source)
triggers a change in a patient (endpoint). Spontaneous changes in (4.10)-(4.11) correspond to the
figure on the right. The role of an external force is conceptually removed; the patient undergoes a
spontaneous change of state with an internal cause rather than external force. A single focused
participant, as source and endpoint for the action, undergoes an internal change of state.84

84

In English and Greek, these kind of patientive intransitive events are common with change-of-state verbs:
bend, break, shatter, burn, die, sink, spill, open. Events like ‘the ice melts’, ‘the glass shatters’, and ‘the leaves burn’,
all involve a patient subject. The event takes place through a spontaneous physical process (via internal means not
external force). These kinds of events form the semantic core for this type of alternation, as presented in figure 4.

94

The schema in Figure 4, contrasting two- and one-participant events, is also applicable to
the next four middle types. The only alteration is the participant role that the single entity plays
in the event. This change in participant role arises from the lexical semantics of the verb. If it is a
mental process, the participant has an experiencer role – the one who undergoes a mental or
emotional change. If the event involves motion, then the single participant plays a mover – the
one who goes through a change in location/posture. This recurring pattern in the Greek middle
accomplishes the same adjustments in focus and salience for body motion, collective motion, and
mental processes, as it does for the physical process type (Maldonado 2008, 165).
4.3.2

Change of state: Body motion

Body motion involves a change in location (leave, return) or a change in body posture (stand, sit,
turn). In an active two-participant event, an agent (source) causes a mover to change
location/posture (endpoint). As energy source and starting point for event construal, the agent is
the external cause that brings about the event. The mover is construed as energy endpoint, the
one that is made to move, as in (4.14) and (4.15).
(4.14) Active caused motion
σὺ ἔστρεψας τὴν καρδίαν τοῦ λαοῦ τούτου ὀπίσω
You turned the heart of this people back (LXX 3 Kingdoms 18:37)
(4.15) ὅτε ἐπλήσθησαν αἱ ἡμέραι τοῦ καθαρισμοῦ αὐτῶν κατὰ τὸν νόμον Μωϋσέως, ἀνήγαγον αὐτὸν
εἰς Ἱεροσόλυμα παραστῆσαι τῷ κυρίῳ
When the days of their purification were complete according to the Law of Moses, they
brought him to Jerusalem to present him to the Lord (Luke 2:22)
A metaphorical change in posture occurs in (4.14); (4.15) references a change in location.
Note ἐπλήσθησαν in (4.15) is a middle-marked verb expressing a spontaneous physical process in
which purification comes to an end (is completed). As a telic event, it expresses a fully achieved
change of state, in keeping with perfective aspect and the historical origins of the -(θ)η- form.

95

In contrast to the active, middle marking expresses a single participant who moves
without external cause. The mover is the primary figure, as both energy source and endpoint. The
energy is internal; the one who induces the change is also the one who undergoes the change, as
in (4.16) and (4.17). The same lexemes from the active events are used, but here the grammatical
subject is the one who experiences the change in location or posture.
(4.16) Middle body motion
ταῦτα εἰποῦσα ἐστράφη εἰς τὰ ὀπίσω, καὶ θεωρεῖ τὸν Ἰησοῦν ἑστῶτα
After saying these things, she turned around and saw Jesus standing there (John 20:14)
(4.17) Ἡμεῖς δὲ προελθόντες ἐπὶ τὸ πλοῖον ἀνήχθημεν ἐπὶ τὴν Ἆσσον
Going on ahead, we set sail for Assos (Acts 20:13)
Because no external force induces the change, no cause/agent subject is needed. The
scope of attention narrows from two participants (active) to one (middle), resulting in greater
salience for the single participant mover. Its syntactic expression reflects this shift in attention.
This middle event appears in an intransitive clause with the mover (object in the active) as
grammatical subject. In some cases, the subject is both an agent who volitionally instigates the
movement and the mover who changes location/posture, as in (4.18). In others, the mover may
be more patient-like without a volitional role. This depends on the lexical semantics of the verb
and its collocation with certain noun phrases, as in (4.19).
(4.18) καὶ πάσης ψυχῆς τῆς κινουμένης ἐν τῷ ὕδατι
and every living thing that moves in the water (LXX Lev. 11:46)
(4.19) ὅπως μὴ κινουμένης ἐμποδίζηται πρὸς τὸ ἔργον, ἀναβαλόμενος ἐπὶ τὸν λαιὸν ὦμον φέρει
that he [the priest] may not be hindered in his work by [his girdle] moving about, he
throws it to the left, and bears it on his shoulder (Josephus, Antiquities 3.155)
Note that middle morphology is used to denote actions that are typically performed by one’s own
bodily efforts, hence ‘body motion middle’ as a descriptive label. A contrasting construal occurs
in (4.20) with active ῥίπτω ‘throw’, a caused motion event that is typically performed on a
distinct patient participant. Here, the verb is used with a reflexive pronoun to signal the non-

96

default case that the action is performed on the agent’s own body rather than on a distinct entity.
Middle morphology is not available for this meaning: ῥίπτομαι *throw oneself.
(4.20) ἵνα μὴ ψαύσειέν τι τοῦ σώματος αὐτῆς, ἑαυτὴν ἔρριψεν κατὰ τῆς πυρᾶς
so that no one would touch her body, she threw herself down into the fire (4 Macc. 17:1)
The typical nature of body motion middles allows for a regular semantic alternation
between causative (active) and anticausative (middle) action. The active widens the scope of
attention for event development, including two onstage participants – agent/source and
mover/endpoint. The middle provides an alternate construal. Removing the external cause
narrows the scope of attention to just the mover – as a natural energy source and endpoint of his
own action. Examples in (4.21) illustrate the regularity of the causative-anticausative pattern in
voice and (4.22) shows middle-only verbs of the same semantic types.
(4.21) Translational motion
Active/Causative
Middle/Anticausative
ἀνάγω
bring up (tr.), moor a ship
ἀνάγομαι
go up, enter a harbor85
ἀπολύω
dismiss, send away
ἀπολύομαι
leave, depart
διαχωρίζω cause distance, separate
διαχωρίζομαι part, leave, go away
ἐκβάλλω drive out, throw out
ἐκβάλλομαι go out, embark
εὐοδόω
lead/help on the way
εὐοδοῦμαι
have a prosperous journey86
κινέω
cause to move, set in motion87 κινέομαι
move (intr.)
περιί̈στημι encircle something
περιί̈σταμαι go around to avoid
πορεύω
make go (Classical)
πορεύομαι
go, travel, walk88
ὑψόω
lift something up
ὑψόομαι
rise up
πλανάω
make wander, make go astray πλανάομαι
wander about, go astray
σῴζω
save, make safe
σῴζομαι
get to safety, escape
Nontranslational motion
Active/Causative
ἀνορθόω
build s.t. up
85

Middle/Anticausative
ἀνορθόομαι
straighten up (intr.)89

In the third person, ἀνάγεται also refers to improvements in the weather, e.g. ‘The storm lifted.’
Primarily occurs with -(θ)η- morphology, metaphorically extends to ‘suceed, prosper’ in one’s actions.
87
Also, non-translational motion sense: cause to shake, shake the head’, e.g. Mark 15:29.
88
Derivative: διαπορεύω ‘carry s.t. through’ (Classical) vs. διαπορεύομαι ‘go through, throughout’.
89
ἐπέθηκεν αὐτῇ τὰς χεῖρας· καὶ παραχρῆμα ἀνωρθώθη ‘he placed his hand on her and immediately she
straightened up’ (Luke 13:13). This illustrates the close relationship between body motion and physical processes.
That both event types are marked by middle morphology is no accident; their relationship is structured and natural.
86

97

κυλίω
μεταβάλλω
μετατίθημι
σείω
στρέφω
τρέπω

roll (tr.)
change, alter, turn
convey, bring, transfer;
shake, agitate (tr.)
cause to turn, return
cause to turn, turn (tr.)

Posture
Active/Causative
βάλλω
put in a location92
ἐγείρω
cause to stand93
ἐπαίρω
lift up, hold up
ἵστημι
place, set, make stand
(καθ)ίζω
cause to sit, seat
κλίνω
cause to bow, bend, fall over
παρακαθίζω make sit near97
(4.22) Media tantum
Translational motion
ἀλάομαι
roam, wander
ἔρχομαι100
come, go, move101
οἴχομαι
go, depart, be gone103
παραλέγομαι sail along past
90

κυλίομαι
μεταβάλλομαι
μετατίθεμαι
σείομαι
στρέφομαι
τρέπομαι

roll about (intr.)
turn oneself, change direction
turn (intr.)
quake, shiver, be agitated
turn, turn around (intr.)90
turn (intr.)91

Middle/Anticausative
βάλλομαι
lie down, lie
ἐγείρομαι
get up, stand up
ἐπαίρομαι
rise up94
ἵσταμαι
stand (intr.)95
(καθ)ίζομαι
sit, take a seat
κλίνομαι
lean, fall, turn96
παρακαθίζομαι sit near98

ἅλλομαι
ἱκνέομαι
παραγίνομαι
πέτομαι

leap, spring up99
arrive at, reach, come102
come, arrive
fly

Derivative: ἀποστρέφω ‘cause to turn away’ vs. ἀποστρέφομαι ‘turn away (intr.)’.
Derivatives: ἀποτρέπω ‘cause to turn away > dissuade’ vs. ἀποτρέπομαι ‘turn away > be dissuaded’;
ἐκτρέπω ‘cause to turn’ vs. ἐκτρέπομαι ‘turn away; be dislocated’; μετατρέπω ‘cause to turn > transform (tr.)’ vs.
μετατρέπομαι ‘turn (intr.) > transform, change (intr.)’.
92
βάλλω has other senses that participate in the active-passive alternation: ‘throw’ vs. ‘be thrown’.
93
Active ἐγείρω may be used intransitively in imperfective aspect commands: Get up! Stand up!
94
The middle ἐπαίρομαι extends metaphorically from ‘rise up’ to ‘be presumptuous’ and ‘put on airs’.
95
The perfect active form ἑστήκα expresses the state that results from the spontaneous (body motion)
process: The standing position (ἑστήκα) is the result of the standing process (ἵσταμαι). Derivatives: ἀνίστημι ‘cause
to stand’ vs. ἀνίστημαι ‘rise up, stand up’; ἀφίστημι ‘remove, draw away, cause to leave’ vs. ἀφίστημαι ‘depart,
withdraw’; ἐφίστημι ‘set, place’ vs. ἐφίστημαι ‘stand near’; καθίστημι ‘set down’ vs. καθίστημαι ‘come before’.
96
Also: ἀνακλίνω ‘cause to recline/lie down’ vs. ἀνακλίνομαι ‘recline to eat’ and κατακλίνω ‘cause to lie
down/sit down’ vs. κατακλίνομαι ‘sit/lie, recline at dinner’.
97
In earlier Greek, παρακαθίζω was an active intransitive ‘sit down’, but by analogy with the larger system
developed a causative sense and in turn, an intransitive non-causal middle.
98
Also: συγκαθίζω ‘cause to sit with’ vs. συγκαθίζομαι ‘sit down with’. Active συγκαθίζω also occurs
intransitively: ‘sit down with’, semantically analogous to the middle.
99
Derivatives with pre-verbs: ἐξάλλομαι ‘leap up/out’; ἐφάλλομαι ‘leap up/out’.
100
Also: ἀπέρχομαι ‘depart’; διέρχομαι ‘go through’; εἰσέρχομαι ‘go into’; ἐξέρχομαι ‘go out’; ἐπεισέρχομαι
‘come upon’; ἐπανέρχομαι ‘go back to’; ἐπέρχομαι ‘come to, arrive’; κατέρχομαι ‘move, come, go down’; παρέρχομαι
‘pass by’; περιέρχομαι ‘wander about’; προσέρχομαι ‘approach, come near’; προέρχομαι ‘pass along’.
101
Note that ἔρχομαι, along with its pre-verb relatives, is active-only in the aorist. See Beekes (2010, 468).
102
Derivatives: ἀφικνέομαι ‘reach, arrive’; διϊκνέομαι ‘move through a space’; ἐφικνέομαι ‘come, arrive’.
103
Derivative παροίχομαι used of time. Time is construed in terms of motion ‘have gone by, be past’.
91

98

Nontranslational motion
ἐπεκτείνομαι reach out, stretch toward104
Posture
ἀνάκειμαι
recline105
(κάθ)ημαι
(καθ)έζομαι
sit down
παράκειμαι
4.3.3

sit down
lie before, be adjacent106

Change of state: Collective motion

A second type of caused motion is collective motion. It occurs when an agent moves multiple
participants en masse. In the active causative construal, the agent as energy source causes a
group to change location as energy endpoint. In (4.23), Joseph is the source and the food that he
gathers is the endpoint. Likewise, (4.24) involves an agent (God) and a mover (people) but in
this case it is two people who are joined together in marriage to become one.
(4.23) Active caused motion
συνήγαγεν πάντα τὰ βρώματα τῶν ἑπτὰ ἐτῶν, ἐν οἷς ἦν ἡ εὐθηνία
he gathered together all the food from the seven years in which there was abundance (LXX
Gen. 41:48)
(4.24) ὃ οὖν ὁ θεὸς συνέζευξεν ἄνθρωπος μὴ χωριζέτω
Therefore, what God has joined no human must separate (Matt. 19:6)
Collective motion expressed using middle morphology is similar, but in (4.25) and (4.26)
no external agent brings about the collective action. The motion is conceptualized as taking place
spontaneously by internal energy. The focus of attention is on what happens to the crowd,
downplaying the salience of an external force by omitting any kind of agent in the syntactic
expression. The secondary participant from the caused active becomes the primary figure put
into profile in the collective middle, making the mover the focal point of event conception.
(4.25) Middle collective motion
Καὶ διαπεράσαντος τοῦ Ἰησοῦ ἐν τῷ πλοίῳ πάλιν εἰς τὸ πέραν συνήχθη ὄχλος πολὺς ἐπʼ αὐτόν
And after Jesus had crossed over in the boat to the other side, a large crowd gathered to
him (Mark 5:21)
104

Via metaphorical extension, ἐπεκτείνομαι has the additional sense ‘to exert oneself to the uttermost’.
Relative to cultural knowledge ἀνάκειμαι extends metaphorically to ‘recline to eat, have dinner’. As do
κατάκειμαι ‘lie down, recline for a meal’ and συνανάκειμαι ‘recline, sit for a meal’.
106
Derivative of κεῖμαι ‘be in a place, lie’: παρά + κειμαι ‘lie before’ > be at hand, ready’ > ‘press, urge’.
105

99

(4.26) ἐγένετο δὲ παροξυσμὸς ὥστε ἀποχωρισθῆναι αὐτοὺς ἀπʼ ἀλλήλων
There was a sharp disagreement, so they separated from each other (Acts 15:39)
The majority of collective motion middles are verbs with preverbs derived from more
basic forms, as in (4.27). The preverb συν- is especially, and predictably common, with its
commitative/associative semantics. The nature of the middle event type, as involving participants
that move collectively in some way, means that most lexemes fall into construals involving
gathering/separating. Variation among them involves the nature of the gathering as a disparate
cluster (ἐπισυνάγομαι) or for a conspiracy (συστρέφομαι). A dispersal might be erratic and
unpredictable (διασκορπίζομαι) or clear and well-defined (ἀποχωρίζομαι). Note the shared
similarities and overlap with the physical process domain (cf. §4.3.1). These are not discrete or
distinct middle types, but clusterings of groups of lexemes with a shared construal: the event
develops from energy that is internal to the process itself.
(4.27) Active/Causative
ἀθροίζω
muster, pull together
ἀποχωρίζω make separate
διαλύω
disperse a crowd
διανέμω
to distribute
διασκορπίζω scatter
ἐπισυνάγω collect
κολλάω
glue, join together
μίγνυμι
bring together, mix
προσκληρόω allot, assign together
προσκλίνω make lean against
σκορπίζω
cause to scatter
συνάγω
bring together
συναθροίζω pull together as a mass
συναλίζω
bring together
συστρέφω rally, gather tightly
συλλέγω
scrape together

107
108

Middle/Anticausative
ἀθροίζομαι
gather for battle/for a unified goal
ἀποχωρίζομαι become separate
διαλύομαι
disperse (intr.)
διανέμομαι
spread out (intr.)
διασκορπίζομαι be scattered
ἐπισυνάγομαι gather unprompted, cluster together
κολλάομαι
join oneself to, cling to
μίγνυμαι
mingle among107
προσκληρόομαι attach oneself to another
προσκλίνομαι join with (+DAT), stand beside
σκορπίζομαι
disperse, scatter
συνάγομαι
assemble, gather together108
συναθροίζομαι gather as a unified group
συναλίζομαι
eat together, stay with
συστρέφομαι gather closely, join a conspiracy
συλλέγομαι
gather (from disparate places?)

Also: συναναμίγνυμι ‘mix up together’ vs. συναναμίγνυμαι ‘mingle, associate with’.
Also: συναπάγω ‘take, lead away’ vs. συναπάγομαι ‘associate with’.

100

Media tantum – Collective motion
συμπορεύομαι
assemble, move as a group
συνέρχομαι
meet together, travel together109
(συν)έπομαι
accompany
συμπαραγί(γ)νομαι come together, arrive together, be ready at the same time
ἐπαθροίζομαι
gather in increasing numbers
4.3.4

Change of state: Mental process

In the active transitive prototype, an agent (source) causes a change in the patient (endpoint).
Among mental events, the same relation holds, but it is metaphorically extended from the
physical to psychological (cognition/emotion) domain. In (4.28)-(4.29), an agent causes a mental
experience/change in an experiencer. In (4.28), the Jews (source) mentally stir up the crowd
(endpoint), causing a change in their state of mind.
(4.28) Active caused experience
οἱ ἀπὸ τῆς Ἀσίας Ἰουδαῖοι θεασάμενοι αὐτὸν ἐν τῷ ἱερῷ συνέχεον πάντα τὸν ὄχλον
The Jews from Asia seeing him in the temple stirred up the entire crowd (Acts 21:27)
(4.29) καὶ μὴ λυπεῖτε τὸ πνεῦμα τὸ ἅγιον τοῦ θεοῦ
And do not grieve the Holy Spirit of God (Eph. 4:30)
The use of middle morphology results in a non-caused (or anticausative) mental process.
These middle events express a spontaneous mental change of state in the experiencer. By shifting
the focus to the experiencer and omitting an explicit agentive external force, attention is focused
on a single participant going through a mental change.
(4.30) Middle mental process
γενομένης δὲ τῆς φωνῆς ταύτης συνῆλθε τὸ πλῆθος καὶ συνεχύθη
At this sound, the crowd came together and grew confused (Acts 2:6)
(4.31) ἐλυπήθησαν σφόδρα
They were extremely depressed (Matt. 17:23)

109

With the addition of συν-, body motion middles like πορεύομαι ‘walk’ and έρχομαι ‘go’ become
collective motion middles ‘walk/go together as a group’.

101

In (4.30)-(4.31), the experiencer is brought forward into the spotlight by profiling it as the
source or starting point for the action. The event occurs not as an external cause, but as an
internal change. The experiencer alone is in view.
From physical processes to motion and mental processes, the use of middle morphology
reflects a change in event construal, specifically a conceptual shift away from the transitive
prototype in which an agent (source) does something to cause a change in a patient (endpoint).
What motivates middle marking in each of these event types is a deviation from the prototypical
active with respect to two parameters: (1) by removing the role of the external agent (causer), the
middle conflates the roles of energy source and endpoint onto a single participant. The energy for
the event is internal to the changed participant. And (2), this narrows the scope of attention.
Rather than two on-stage participants (primary and secondary figures) there is just one. Attention
naturally falls onto the single participant, i.e. the one undergoing some kind of change – be it
physical state/property, mental experience, or motion.
Each syntactic expression reflects conceptual content. In the active caused event, the
experiencer/mover/patient is expressed as syntactic object – the secondary downstream figure in
the event. But in the anticausative middle, these same participants become the syntactic subject –
as primary figures. In the resulting intransitive predication, the subject does not play the role of
external cause as in the active, but plays the role of experiencer, mover, or patient. Their
participant roles do not change, only their heightened salience in event conception.
Of course, there are alternate means of coding psychological events. Each formal strategy
embodies a different type of event construal, imposing different imagery on event development.
Some of these alternate means are discussed in (4.33) below. Others are discussed in §4.4.1.

102

Causative/anticausative mental processes are listed in (4.32). Additionally, many mental
processes are expressed via metaphor from cultural and embodied domains of experience. Active
παχύνω may mean ‘cause to be fat, fatten’ but also ‘cause to be mentally dull/calloused’. The
former refers to a caused physical state, the latter a psychological one.
(4.32) Active/Causative
αἰσχύνω
cause shame110
διακρίνω
make a distinction, judge
112
διαταράσσω
cause confusion, make upset
ἐκθαμβέω113 cause to be amazed
ἐκλανθάνω
escape notice, make forget
ἐμπίπλημι
fill, satisfy
114
ἐνοχλέω
cause discomfort, annoy
ἐξίστημι
amaze, confuse, displace
εὐφραίνω
make glad
ἥδω (rare)
please, make glad
θαμβέω115
cause astonishment
116
θυμόω
make angry
κοιμάω
put to sleep
λυπέω
make sad
μωραίνω
make foolish, remind
φοβέω118
cause to be afraid
φράζω
make known, show forth
φυσιόω
inflate, make proud
ξενίζω
cause surprise
119
παχύνω
make dull
πείθω
convince, persuade
ὑπεραίρω
make high, lift
(ὑπο)μιμνῄσκω cause to remember

110

Middle/Anticausative
αἰσχύνομαι
be/feel ashamed
διακρίνομαι
have doubt111
διαταράσσομαι
become confused, upset
ἐκθαμβέομαι
be amazed
ἐκλανθάνομαι
forget
ἐμπίπλημαι
become full > enjoy
ἐνοχλέομαι
become annoyed
ἐξίστημαι
be amazed, astonished
εὐφραίνομαι
become glad
ἥδομαι
enjoy oneself, delight in
θαμβέομαι
be astonished
θυμόομαι
become angry
κοιμάομαι
fall asleep117
λυπέομαι
become sad
μωραίνομαι
become foolish
φοβέομαι
become afraid, fear
φράζομαι
ponder, think, perceive
φυσιόομαι
become inflated, proud
ξενίζομαι
become surprised
παχύνομαι
become dull
πείθομαι
believe, be certain, obey
ὑπεραίρομαι
exalt oneself, be elated
(ὑπο)μιμνῄσκομαι remember

See also καταισχύνω.
That is: ‘be divided psychologically/mentally’.
112
See also: θορυβέω/θορυβέομαι, θορυβέω/θορυβέομαι, παροξύνω/παροξύνομαι, and συγχέω/συγχέομαι.
113
See also: θαμβέω/θαμβέομαι.
114
See also: σκύλλω/σκύλλομαι.
115
New causative back-formation; the older active form was intransitive and still occurs in some authors.
116
See also: μαίνω/μαίνομαι, ὀργίζω/ὀργίζομαι, and πυρόω/πυρόομαι, which involve a metaphorical
extension via the metaphor ANGER IS HEAT: burn > upset.
117
This verb also extends metaphorically to sexual intercourse ‘lie with’ and death ‘fall asleep > die’.
118
See also: πτοέω/πτοέομαι and πτύρω/πτύρομαι.
119
This verb also has a causative vs. physical process alternation: ‘make fat’ vs. ‘become fat’.
111

103

An additional set of mental processes in (4.33) do not necessarily entail a causativeanticausative contrast (though some do), but instead express a physical-psychological distinction.
A basic physical agent-patient relation in the active is mapped to an experiencer-stimulus
relation in the middle.120 One instance of a caused event is καταλαμβάνω ‘make s.t. one’s own,
take hold of’. The active is used in contexts of pursuit and capture. Often, the middle supplies a
passive contrual; the object that is seized becomes the subject and the agent is demoted. But for
those middles that express a mental process, there is no demotion of the agent. Rather, the event
of pursuit and attainment is mapped onto the domain of cognition: ‘I grasped (κατελαβόμην) that
he had done nothing worthy of death’ (Acts 25:25) or ‘grasping (καταλαμβανόμενος) in his mind
the clever tricks she put forward’ (Josephus Antiquities 8.167). The learning process is mapped
onto pursuit, and comprehension onto attainment of an idea. English verbs share a similar
mapping: Leah grasped the concept quickly or Emily seized on the idea with enthusiasm.121
Other verbs do not involve caused events, e.g. ἀρκέω in (4.33). Note that its argument
realization reflects differences in conceptual event structure. The active form usually takes a
dative oblique argument denoting a beneficiary and refers to a situation in which the relative
state of an object (a condition/quantity) is deemed (in-) sufficient: ‘Two hundred denarii of bread
would not be enough (ἀρκοῦσιν) for them’ (John 6:7). The middle reconstrues the event from the
view of the experiencer/recipient (as subject) and, if it occurs at all, the physical object is
expressed as an oblique dative: ‘Be content (ἀρκεῖσθε) with your pay’ (Luke 3:14).
(4.33) Active/Physical
ἀρκέω
s.t. is sufficient/adequate
120

Middle/Psychological
ἀρκέομαι
be satisfied/content

For discussion of experiencer and stimulus roles, see §4.4.1.
A common construal across languages: Experience of physical control over objects is mapped onto
mental manipulation. Physical objects that we control are those within our reach. Mental concepts that we control
are those that we can process and understand (in contrast to ideas that we do not ‘have hold of’) (Sweetser 1990, 38).
121

104

ἀνέχω
ἀντέχω
διαπονέω
διαπρίω
ἐκκρεμάννυμι
ζημιόω
καταλαμβάνω
σαλεύω

make erect, lift up, prop up122
hold against, hold out
cultivate, till a field
saw/cut through124
cause to be suspended, hang
cause injury
seize, take hold of, grasp
cause to shake

ἀνέχομαι
hold oneself up, endure
ἀντέχομαι
cling to, resist
διαπονέομαι
be troubled123
διαπρίομαι
become infuriated
ἐκκρέμαμαι
cling to > attend to
ζημιόομαι
experience hardship
καταλαμβάνομαι comprehend
σαλεύομαι
feel shaken (mentally)

Many verbs, whether the active is causative or not, often maintain a separate middle
usage that is not a mental process, but fits another middle type, e.g. body motion. There is a
tendency for a more regular alternation of the primary sense of the verb alongside a secondary
psychological sense that is middle only. Regardless, all involve a metaphorical construal of the
mental domain in terms of a physical interaction. Middle-only mental processes are in (4.34).
(4.34) Media tantum – Pyschological
ἄγᾰμαι
admire, wonder
ἀνδρίζομαι
be courageous
ἄχθομαι
be vexed, be angry with
διανοέομαι
think, intend, understand
ἐμμαίνομαι
become enraged
ἐπαισχύνομαι be ashamed, feel shame127
ἐπιλανθάνομαι forget
ἐπίσταμαι
understand, be acquainted
εὐλαβέομαι
be concerned, dread, fear
μετεωρίζομαι be anxious about
ὁμείρομαι
have affection, yearn for
σέβομαι
revere, worship129
συνήδομαι
delight in

αἰδέομαι
ἀπεκδέχομαι
βούλομαι
(δι)ενθυμέομαι
ἐνυπνιάζομαι
ἐπαναπαύομαι
ἐπιμελέομαι
ἔρᾰμαι
μεταμέλομαι
οἴομαι
προσδέομαι
σπλαγχνίζομαι
χρηστεύομαι

be ashamed, feel regard for
wait eagerly
want, desire, intend, plan125
ponder, contemplate
dream126
rely on, trust, rest upon
attend, take care of, manage
love, lust, desire128
regret, change one’s mind
think, suppose, imagine
be in need of, want
feel compassion
be kind, show kindness

Each anticausative middle type (physical process, body motion, collective motion, and
mental process) provides a new starting point or energy source for event conception. Source and
122

Also, active intransitive: rise up, emerge.
This verb involves a metaphoric extension where one’s mental state is construed as tilled up like a field.
124
Metaphorical extension: ‘cause emotional pain’.
125
Less volitional senses: ‘want, desire’, more volitional senses: ‘plan a course of action, intend to do’.
126
This verb was in earlier history an activa tantum that changed into a media tantum.
127
Later Byzantine Greek develops a causative active form with the sense 'make ugly'.
128
This is a middle alternative to the activa tantum: ἐράω. See also Beekes (2010, 449).
129
Also: σεβάζομαι ‘show reverence, worship’.
123

105

endpoint roles are conflated onto the same participant with internal energy supplying the energy
transfer for the event (rather than external energy in the active). For both the passive and
anticausative types there is a limited to vanishing saliency for an external source of energy
(Kemmer 1993, 205). Among anticausatives, removing the external cause provides a shift in
attention toward the participant that is internal to the core process. This narrows the scope of
attention and heightens the salience of the change undergone by the single participant.
4.4

Alternations in voice: Symmetrical energy in cognitive events

As noted in §4.3.4, cognitive events vary in their formal expression, with each coding strategy
representing a different event construal. In (4.35), the mental event in (a) is coded much like the
transitive prototype, with active morphology. The stimulus, as external source/cause for the
mental event, is coded as primary figure (subject), with the experiencer, the changed participant,
as secondary focal point (object). The opposite realization is in effect with middle morphology in
(b). The changed participant (experiencer), as primary focal point is realized as subject, with the
stimulus, source/cause for the change, coded as object. In (c), the external cause is removed, and
the experiencer serves as a single focused participant.
(4.35) (a) ἐφόβησε αὐτὸν ὁ ὄχλος the crowd frightened him
(b) ἐφοβήθη τὸν ὄχλον
he feared the crowd
(c) ἐφοβήθη
he became afraid

active (stimulus → experiencer)
middle (experiencer → stimulus)
middle (experiencer)

One reason for this type of variation in coding is that cognitive events (unlike physical
ones) involve a symmetrical, or bidirectional transmission of force, allowing either experiencer
or stimulus to be coded as starting point (source) or endpoint (effect) in event conception. In a
two-participant mental event, an experiencer directs attention to a stimulus. That’s one relational
arc. The stimulus, in turn, produces some sort of mental change/response in the experiencer.

106

That’s the second relational arc (Croft 2012, 233). Symmetrical energy in cognitive events
represents a contrast with the transmission of force in prototypical transitives.
In the transitive prototype, one participant supplies the energy source, exerting force on a
secondary figure that receives the energy, as endpoint, and is affected by it. In a force-dynamic
model of event conception, causal relationships are construed as directed and asymmetric. Based
on a general metaphorical mapping, experience of a spatial path from source to goal is mapped
onto transmission of force; energy flows in one direction along a path from source/cause to
endpoint/effect. Yet, as illustrated in (4.35), there are circumstances in human experience, like
mental events, that involve symmetrical rather than asymmetrical energy flow.
Symmetrical energy transfer is present among cognitive events in the middle domain, i.e.
those that imply mental attention of one sort or another, as in mental processes (§4.3.4), mental
activities (§4.4.1), speech acts (§4.4.2), and perception (§4.4.3). Transmission of force is
bidirectional, with two participants and two relations. An experiencer directs attention to, or
becomes aware of, a stimulus. The stimulus, in turn, engenders a cognitive change in the mind of
the experiencer. In middle expressions, the primary figure (experiencer) is a sentient participant
in whose mind a mental event occurs. Depending on the verb, this may be a mental activity (e.g.
devising a plan), a perceptual experience (e.g. visual/tactile observation), or a speech act that
expresses something about the mental state or thoughts of the speaker. The stimulus, as
secondary figure, may be a real-world entity or an imagined representation thereof, as with a
salient concept or other perceptual prompt (Kemmer 1993, 127; Croft 2012, 221-33).
An alternative construal (common for mental processes in §4.3.4) involves just one
participant with no stimulus coded in syntactic structure. This type of one-participant construal
may be preferred if there is no relevant external stimulus that occasioned the event, such as when

107

the stimulus is understood to be solely internal to the experiencer. One-participant relations
function to pragmatically downplay or remove the role of any external stimulus so that the event
transpires exclusively with regard to its effect on the experiencer. This alternation among mental
events is similar to that among physical processes and motion in §4.3.1 -§4.3.3. Removing an
external cause has the effect of placing the changed participant into the spotlight.
The formal expression of mental events reflects a semantic-pragmatic motivation. For
two-participant mental events, middle morphology signals a deviation from the basic agentpatient interaction in the active transitive prototype. In the middle, the experiencer, unlike a
typical agent, is both energy source and ultimate endpoint for the relational arc of the event. Two
roles are conflated onto a single participant; the experiencer is both the initiating force and the
participant that experiences the endpoint change. For one-participant events, the intransitive
nature of the predication focuses attention on a single participant that is source and endpoint,
downplaying or eliminating the role of any secondary figure.
For mental construals, the experiencer is at both ends of a relational arc or bidirectional
energy transfer. The experiencer is the energy source in whose mind the mental event originates
and the energy endpoint as the one who is cognitively changed by the experience. The stimulus,
as the object of perception, unlike a prototypical patient, remains unaffected by the interaction
(Kemmer 1993, 127-129; Kemmer and Verhagen 1994; Croft 2012, 221-33).
4.4.1

Mental activity

Middle event types fall on a continuum from those that are more patient-like and lower in
transitivity to those that are more agent-like and higher on the transitivity scale. The shift from
mental processes (§4.3.4) to other cognitive events represents a subtle part of this transition.
Verbs of cognition, as a whole, are lower on the transitivity scale since they do not express a

108

typical agent-patient interaction, one in which a volitional agent instigates an action to cause a
physical change of state in a distinct patient. Yet cognitive events do still vary in their degree of
transitivity as well as in the nature of energy transfer in a given cognitive predication. Mental
processes entail lower transitivity with a more patient-like primary figure that undergoes but
does not generally instigate the cognitive change. This represents a shift for the next set of
cognitive events (mental activity/speech act/perception), where the experiencer plays a volitional
role in instigating the event with a particular goal in mind, resulting in a more agent-like primary
figure that still experiences the mental effect of the action.
The line of demarcation between the two (mental activity vs. mental process) is not an
exacting division but runs along a scale of volitionality. Humans do not always have control over
where our attention is directed. Mental processes tend to describe events in which a primary
figure is less volitional and more patient-like, whereas mental activities tend to describe events in
which a primary participant acts volitionally as an agent-like subject. Their morphological
expression reflects this semantic difference. As illustrated in (3.20)-(3.21) in §3.2.2, mental
processes, as more patient-like events, receive -(θ)η- morphology in the perfective paradigm. In
contrast, volitional mental activities primarily take the -σα- middle form. Examples in each
section (contrast mental processes (4.30)-(4.31) and mental activities (4.36)-(4.37)) reflect this
formal shift in middle morphology between -(θ)η- and -σα- forms.
In a two-participant mental activity, an experiencer directs his mental attention to a
stimulus. In turn, a mental event is brought about in the mind of the experiencer, so that he is
cognitively changed in the course of the event (Kemmer 1993, 129).

109

Figure 5

Two-participant mental activity

Although there are two participants, this event type still departs from the active transitive
in that it profiles a change not in the secondary figure, but in the primary participant. In the
transitive prototype, the agent causes a change to take place in the patient; no profiled change
occurs for the agent. But for mental events, the opposite is true. The profiled change occurs in
the primary figure (experiencer) instead. The stimulus, as secondary figure, remains unchanged.
In (4.36), Tobit keeps track of the days, anxiously awaiting the arrival of his son. And Solomon,
in (4.37), applies mental effort in reflecting on the judgments of God. Note the use of the
perfective -σα- middle form with mental activities (rather than -(θ)η- with mental processes).
(4.36) Mental activity middle
Καὶ Τωβεὶτ ὁ πατὴρ αὐτοῦ ἐλογίσατο ἑκάστης ἡμέρας
Now his father Tobit counted each day (LXX Tobit 10:1)
(4.37) Ἀνελογισάμην τὰ κρίματα τοῦ θεοῦ ἀπὸ κτίσεως οὐρανοῦ καὶ γῆς
I considered the judgments of God from the creation of sky and earth (LXX Ps. of Sol. 8:7)
For one-participant mental activities, a single participant goes through a mental change.
These cognitive activities are much like one-participant mental processes, but the experiencer
subject is generally more volitionally involved as an instigator of the event.

Figure 6

One-participant mental activity

The dotted arrow in Figure 6 indicates that with mental activities no distinct secondary
figure is changed by the action. This also reflects its deviation from the active transitive in which

110

an agent causes a change in a distinct patient. In mental activity middles, the change caused by
the event does not alter a distinct entity but the one who initiated it. Rather than two participants
sharing the stage, the spotlight is focused on one figure, serving as primary focal point for event
development. The experiencer is mentally altered by his involvement. The -σα- middle in (4.38)
is used in an intransitive predication; the primary figure is mental source and altered endpoint.
(4.38) καὶ ἐβουλεύσατο ὁ βασιλεύς
And the king deliberated (3 Kingdoms 12:28)
Like mental processes, mental activities may also arise via metaphoric extension based
on other event types. The verb στρέφω is used as as a body motion middle: ‘turn’. The active
expresses caused motion: ‘turn s.o./s.t. around.’ The middle expresses an internally-induced
single participant construal: ‘turn around’, as in ‘do not throw your pearls before swine, or they
will trample them under foot and turn (στραφέντες) and maul you’ (Matt. 7:6).
Basic body motion is then extended from the physical to psychological domain, so
στρέφω also fits among cognitive middles: ‘Truly I tell you, unless you turn (στραφῆτε) and
become like children, you will never enter the kingdom of heaven’ (Matt. 18:3). Its use is more
about a mental attitude than a physical turn. A single lexeme fits both motion and cognition. This
kind of lexical polysemy allows for metaphorical extension throughout the middle network.
Mental activities display many of the same alternation patterns as mental processes (§4.3.4),
though see the discussion of the causative/anticausative pattern below.
(4.39) Active/Causative
βδελύσσω make detestable
σοφίζω
make wise, instruct

130

Middle/Anticausative
βδελύσσομαι regard with disgust, detest
σοφίζομαι
reason out, devise, concoct130

Both βδελύσσομαι and σοφίζομαι are derivatives (βδελυρός ‘disgusting’; σοφός ‘skillfull’/σοφία
‘skillfulness’). The middle is the older expression, with the active form as a later Hellenistic addition, fitting such
verbs within the larger causative/anticausative voice pattern (Beekes 2010, 208, 1374).

111

Active/Physical
μεταβάλλω cause to change, alter
ὀρέγω
reach out for s.t.
προαιρέω bring s.t. forth, take out
προβλέπω foresee, see beforehand
σημειόω
make a mark on, designate
σταθμάω measure with a rule/line
τίθημι
lay, put, set up
Media tantum – Psychological
ἀναλογίζομαι consider, reason, think
βουλεύομαι deliberate, devise
ἐγκρατεύομαι exert self-control
λογίζομαι
calculate, evaluate, consider
μιμέομαι
imitate, emulate, follow
στοχάζομαι calculate, survey, explore

Middle/Psychological
μεταβάλλομαι change one's purpose/mind
ὀρέγομαι
aspire to, desire
προαιρέομαι determine, decide, prefer
προβλέπομαι select in advance, provide for
σημειόομαι
take note, pay attention to
σταθμάομαι estimate, assess, measure
τίθημαι
contrive, resolve, effect s.t.131
ἀποδέχομαι
διαλογίζομαι
ἡγέομαι
μηχᾰνάομαι
παραιτέομαι
τεχνάομαι

recognize, approve of
consider, reason, discuss
think, consider
contrive, devise, plot132
excuse, reject, avoid
contrive, craft

For anticausative alternations in (4.39), there is a semantic difference in the use of -σαand -(θ)η- morphology, illustrated with perfective aorist forms in (4.40).
(4.40) Active/Causative
ἐβδελύξατε make detestable
ἐσόφισεν

Middle/Anticausative
ἐβδελύχθη become detestable/loathsome
ἐβδελύξατο regard with disgust, detest, abhor
make wise, instruct ἐσοφίσθης become wise
ἐσοφίσατο gain wisdom, reason out, devise, concoct

Active ἐσόφισεν expresses a caused mental change ‘make s.o. wise’, whereas the middle
removes the external agent as the source of energy for the event so that the experiencer becomes
the primary mental source for the change. The -σα- form in (4.41) expresses a volitional mental
activity, whereas -(θ)η- in (b) expresses a less volitional mental process. The causativeanticausative pattern tends to be more robust among change-of-state verbs, where the contrast
between high transitivity in the active (an agent causes a change of state in a patient) and low
transitivity in the middle (a patient undergoes a change of state) is most evident.

131

Derivatives: μετατίθημι ‘bring, transfer’ vs. μετατίθεμαι ‘change one’s mind, turn away’; συντίθημι ‘put
with’ vs. συντίθημαι ‘agree, decide, affirm’; συγκατατίθημι ‘deposit together’ vs. συγκατατίθεμαι ‘agree with’.
132
Also, μήδομαι ‘plan, contrive’ and μητίομαι ‘devise’.

112

(4.41) (a) καὶ ἐσοφίσατο Ηλι ὅτι κύριος κέκληκεν τὸ παιδάριον
And Eli reasoned out that the Lord had been calling the child (1 Kingdoms 3:8)
(b) καὶ ὁ ἐλασσούμενος πράξει αὐτοῦ σοφισθήσεται
and the one who reduces his activity will become wise (Sirach 38:24)
A verb like ἱλάσκομαι shows a similar semantic difference with -σα- and -(θ)η- forms. The
-σα- middle is used of men in their relation to deities, expressing appeasement of the gods,
imploring their favor, appeasing their anger, and appealing to their mercy. This is often
accompanied by sacrifice, as with seeking a pardon for sin in (4.42). The -(θ)η- form in (b) is
used of deities in their relation to men, expressing pardon for sin, the granting of favor, and the
giving of mercy. Thus, the -σα- middle represents propitiation of the gods, with the -(θ)η- middle
used to denote the gods becoming propitious toward men.
(4.42) (a) πρότερον γὰρ εὐξάμενοι καὶ θυσίας ἀναγαγόντες καὶ ἱλασάμενοι τὸ θεῖον
For after having first prayed and presented sacrifices and implored the favour of the
deity (Philo Planting 162)
(b) καὶ ἱλάσθη Κύριος περιποιῆσαι τὸν λαὸν αὐτοῦ
And so the Lord was favorably inclined to preserve his people (Exodus 32:14)
Two further types in the mental domain include speech acts and perception middles. Both
involve the interplay of experiencer and stimulus roles. The schema in Figure 5 above pertains as
much to speech acts and perceptions as it does to mental activity middles. In the perception and
speech act types, similar patterns are repeated as in mental activities. The only difference is the
lexical semantics of the verbs: speech and perception (Kemmer 1993, 133-36).
4.4.2

Speech act

Speech acts denote a class of verbs that relate to communication and verbal exchange. Some
describe the transfer of information (ἀνατίθημαι ‘explain’) or conversational interchange
(ἀποκρίνομαι ‘answer, reply’). Others specify something about the mental attitude or
communicative purpose of the speaker (περπερεύομαι ‘brag’; μέμφομαι ‘blame’). Though they

113

may differ in how a message is communicated (παρρησιάζομαι ‘speak boldly’; προκαλέομαι ‘call
out angrily’), they share a similar conceptual structure (Levin 1993, 202).
In a two-participant speech act, an experiencer, as primary figure, directs attention to a
secondary stimulus (real or imagined). And by virtue of the interaction, the stimulus brings about
a cognitive, and indeed expressive, reaction on the part of the experiencer. Middle marking is
used to signal an explicit difference among speech acts from event development in a prototypical
agent-patient interaction. Instead of profiling a change in a distinct endpoint, speech acts portray
an event in which an experiencer is both source and endpoint of a relational arc. Conflating these
roles into a single participant (rather than two distinct participants) highlights the speaker’s
experience and apprehension of the stimulus (Langacker 2006, 128-30).
(4.43) Speech act middle
καὶ προσκαλεσάμενος αὐτοὺς ὁ Πιλᾶτος λέγει· εἴπατέ μοι ὅτι πῶς δύναμαι ἐγὼ ἡγεμὼν ὢν
βασιλέα ἐξετάσαι
And Pilate summoned them and said: Tell me, how can I, as a governor, examine a king?
(The Acts of Pilate 1:2)
(4.44) καὶ προσηυξάμην πρὸς Κύριον τὸν θεόν· ἐξωμολογησάμην
And I prayed to the Lord God and made confession (LXX Daniel 9:4)
Some speech acts are similar to benefactive middles (see §4.5.3). Certain events confer
an additional benefactive or recipient role onto the experiencer, as with Pilate summoning the
Jews to himself in (4.43). Speech acts like εὔχομαι ‘pray’, καυχάομαι ‘boast’, περπερεύομαι
‘brag’, and ψεύδομαι ‘lie’ are examples of such types, as are ἀπολογέομαι ‘speak in one's defense’
and διακατελέγχομαι ‘refute in debate’. In (4.45), false prophets gain benefit by lying to the
people. Their contrition comes as a result of their exposure as false prophets. Note the oneparticipant intransitive construal. No secondary stimulus is syntactically coded since it can be
inferred from context that they lied about the Lord to the people.
(4.45) καὶ ἐνδύσονται δέρριν τριχίνην ἀνθʼ ὧν ἐψεύσαντο
and they [false prophets] will put on a cloak of hair, because they lied (Zech. 13:4)

114

Voice alternations among speech acts, in (4.46), tend to be idiosyncratic and depend on
the lexical semantics of a given verb. Often this involves an alternation in event development,
especially in regard to the construal of energy endpoint: ἐπαγγέλλω ‘command someone to do
something’ vs. ἐπαγγέλλομαι ‘promise to do something oneself’; συμβουλεύω ‘advise, give
counsel to’ vs. συμβουλεύομαι ‘take counsel, consult, plot’. Other alternations may rely on a
difference between a non-speech act in the active (ἀνατίθημι ‘lay s.t. out’) vs. a speech act in the
middle (ἀνατίθημαι ‘lay out in speech, explain’). If lexical senses are not fully distinguished
between active and middle forms, then some semantic overlap is to be expected, as in (4.48).
(4.46) Active/Physical
ἀνατίθημι put, lay s.t. out
ἀποτάσσω set apart
ῥώννυμι
make strong
Active/Speech act
ἐξομολογέω agree, accept (rare)
ἐπαγγέλλω give orders
παρακελεύω command, order
συμβουλεύω advise, give counsel to
ψεύδω
deceive, cheat, disappoint

Middle/Speech act
ἀνατίθημαι
impart, communicate, lay out
ἀποτάσσομαι part with > bid adieu
ῥώννυμαι
be in good health > bid farewell
Middle/Speech act
ἐξομολογέομαι confess, admit, acknowledge133
ἐπαγγέλλομαι promise, profess, claim
παρακελεύομαι exhort, encourage
συμβουλεύομαι consult/ask, plot with
ψεύδομαι
lie, speak falsely

(4.47) Media tantum – Speech act
ἀναίνομαι
refuse, reject
(ἀπ)αρνέομαι
deny134
ἀπολογέομαι
speak in one's defense
δέομαι
plead, beg, ask
διακατελέγχομαι refute in a debate
διηγέομαι
inform, relate, tell136
εἰσκαλέομαι
invite in
ἐρεύγομαι
blurt out137
133

ἀνθομολογέομαι
ἀποκρίνομαι
(ἀπο)φθέγγομαι
διαβεβαιόομαι
διαμαρτύρομαι
διϊσχυρίζομαι
ἐμβριμάομαι
εὔχομαι

give thanks
answer, reply135
speak, utter, speak out
insist, state confidently
solemnly urge
insist, maintain firmly
insist, warn sternly
speak to God, pray138

The middle is the older form with the active ἐξομολογέω as a later addition in Hellenistic, perhaps by
analogy to a more longstanding active form with a similar meaning: ὁμολογέω ‘agree to, admit’.
134
The verb ἀπαρνέομαι may have the additional sense: ‘act selflessly,’ i.e. deny one’s self.
135
Also: ἀνταποκρίνομαι ‘answer, reply in opposition’.
136
Also: ἐκδιηγέομαι ‘inform, relate, tell fully’; ἐξηγέομαι ‘make known, tell fully’.
137
The sense: ‘blurt out’ arises via metaphor from ‘belch, disgorge’. Forceful discharge out of the mouth is
extended to the domain of speech as forceful expression.
138
Also: προσεύχομαι ‘pray’.

115

(κατα)καυχάομαι
λίσσομαι
μαρτύρομαι
μυκάομαι
ὀδύρομαι
παραιτέομαι
παρρησιάζομαι
προεπαγγέλλομαι
προμαρτύρομαι
προσκαλέομαι
πυνθάνομαι

boast139
beg, pray, entreat
affirm, testify
bellow, roar141
mourn, lament
ask for, entreat
speak boldly
announce before142
predict, foretell
summon, call to oneself
inquire, ask, learn about

(4.48) Formal variation – Speech act
αἰτέω/ομαι
ask, demand, plead
διακρίνω/ομαι
judge, criticize
(δια)τάσσω/ομαι give instruction, order
ἐξαιτέω/ομαι
demand, request
εὐαγγελίζω/ομαι proclaim good news143
μετακαλέω/ομαι summon144
παραγγέλλω/ομαι order, command
προστάσσω/ομαι instruct, determine

(κατ)αράομαι
μαντεύομαι
μέμφομαι
μωμάομαι
παραβιάζομαι
παραμυθέομαι
περπερεύομαι
προκαλέομαι
προοιμιάζομαι
προφασίζομαι
ὑπισχνέομαι
ἀπειλέω/ομαι
διαστέλλω/ομαι
ἐντέλλω/ομαι
ἐπικαλέω/ομαι
λοιδορέω/ομαι
ὁμολογέω/ομαι
προσαπειλέω/ομαι
συγκαλέω/ομαι

swear, curse140
tell fortunes
censure, blame
blame
urge strongly
speak soothing, console
brag, be boastful
call out angrily, provoke
make a prelude
make excuses
promise
warn, threaten s.o.
command, assert
instruct, command
call on, invoke, appeal
revile, rebuke
profess, confess
threaten further
call together, summon

As noted in §3.1.1, middle morphology often shows idiosyncratic distribution among
lexical types. This is especially true among speech acts. A verb like θρηνέω ‘wail, lament’ takes
active morphology, but ὀλοφύρομαι ‘lament, wail, moan’ receives middle marking. Idiosyncrasies
like this are the norm among middle-marking languages. This is largely due to its semantic
nature. As middle morphology spreads from one verb to another, idiosyncrasies arise as a natural
consequence of a lexically-driven process.

139

Also: ἐγκαυχάομαι ‘boast’.
Use of the -(θ)η- suffix allows for the passivization of the sigmatic middle.
141
In addition to human speech acts, a number of animal/human sounds also receive middle marking:
βληχχάομαι ‘bleat’, βρωμάομαι ‘bray’, κνυζέομαι ‘wimper’, μηκάομαι ‘bleat’, ὀγκάομαι ‘bray’, φρῑμάσσομαι ‘snort’,
φρυάσσομαι ‘whinny’, χασμάομαι yawn, ὠρύομαι ‘roar, howl’.
142
An active does occur in Dio Cassius Historicus 38.13, canvass for an office before.
143
The middle also takes a passivizing function of the active, ‘be proclaimed [by]’. Derivative:
προευαγγελίζομαι ‘proclaim good news in advance.
144
μεταπέμπω/ομαι ‘send for, summon’ is sometimes active, but mostly middle.
140

116

But there are some lexical tendencies among speech acts in Greek that are worth noting.
General verbs of speech, those that are semantically neutral, often default to active morphology,
e.g. λέγω ‘speak, say’, λαλέω ‘speak, talk’, ἐρωτάω ‘ask’ (Allan 2003, 106). Verbs with more
precise connotations, especially those with an emotional element in their lexical semantics, tend
to receive middle marking: παρρησιάζομαι ‘speak boldly with courage’; διαβεβαιόομαι ‘state
confidently’; παραμυθέομαι ‘speak soothing to’. This also applies to verbs of announcement
(ἀγγέλλω ‘tell’ vs. ἐρεύγομαι ‘blurt out, express forcefully’) and assertion (φάσκω ‘assert, claim’
vs. μαρτύρομαι ‘testify from personal knowledge’). With others, a lexical opposition may be at
play: εὐλογέω ‘bless’ vs. καταράομαι ‘curse’; ἀληθεύω ‘speak truth’ vs. ψεύδομαι ‘speak false’.
Verbs of reciprocal exchange often receive middle marking. Either an experiencer
solicit’s information (πυνθάνομαι ‘inquire, ask, learn’) or imparts information in response to a
previous utterance (ἀποκρίνομαι ‘answer, reply’). The same occurs among verbs that involve
future obligation on the part of the speaker: ὑπισχνέομαι ‘promise’; μαντεύομαι ‘prophesy’. Some
speech acts share a lexical root, which tends to shape their morphological expression; ἡγέομαι
‘consider, regard’ is a mental activity middle, but it also extends to speech acts. An experiencer
makes something known by relating what he has seen/heard: διηγέομαι ‘describe in detail’;
ἐκδιηγέομαι ‘inform fully’; ἐξηγέομαι ‘tell fully’. Similarly, ἀνθομολογέομαι ‘thank’ and
ἐξομολογέομαι ‘confess’ share a lexical root ὁμολογέομαι ‘profess’. All involve an experiencer
who expresses out loud what is already internally acknowledged (Kemmer 1993, 133-41).
4.4.3

Perception

Perception verbs pertain to events involving the perceptual modalities: taste, touch, see, smell,
and hear. In a two-participant interaction, an experiencer directs attention to, or becomes aware

117

of, a stimulus via the senses. As a result, the stimulus induces a perceptual or cognitive encounter
on the part of the experiencer (Kemmer 1993, 136; Croft 2012, 233).
(4.49) Perception middle
ἡμεῖς μὲν τῶν ἡψημένων βρωμάτων παραθήσομεν· σὺ δὲ ὑποκρινόμενος τῶν ὑείων
ἀπογεύσασθαι σώθητι
We will serve you boiled meat. As for you, save yourself by pretending to taste the pig (4
Macc. 6:15)
(4.50) τίς γὰρ ἀνδριάντας ἢ γραφὰς θεασάμενος οὐκ εὐθὺς ἐνενόησεν ἀνδριαντοποιὸν ἢ ζωγράφον;
Who can look upon statutes or paintings without thinking at once of a sculptor or painter?
(Philo The Special Laws I 33)
As with other mental events, the middle form is used among perception verbs to signal an
explicit difference between an experiencer-stimulus relation and a typical agent-patient relation.
Rather than profiling a change in a distinct endpoint participant, perception middles portray an
event that involves a bidirectional transmission of force; an experiencer attends to a stimulus and
the stimulus causes a mental change in the experiencer. The experiencer plays both source and
endpoint roles. Conflating these roles onto a single participant – rather than on two distinct
participants – has the consequence of highlighting the experiencer’s apprehension of, or relation
to, an external stimulus, as illustrated in Figure 7 (Figure 5 in §4.4.1).

Figure 7

Two-participant perception

Two common alternations apply among perception verbs: One is the causative (active)
vs. anticausative (middle) pattern persistent in Greek voice. The other is the distinction between
physical and psychological experience: A physical agent-patient relation in the active alternates
with an experiencer-stimulus relation in the middle. This pattern is common to the cognitive
domain, as noted with mental processes §4.3.4 and mental activities §4.4.1.

118

(4.51) Active/Causative
γεύω
cause to taste
κατοπτρίζω show in a mirror
ὀσφραίνω make smell (rare)
θεάω
cause to see (rare)
Active/Physical
ἅπτω
light a fire, fasten, grasp
ἐφάπτω
bind to, fasten

Middle/Anticausative
γεύομαι
taste145
κατοπτρίζομαι look in a mirror, reflect
ὀσφραίνομαι smell
θεάομαι
observe, look at
Middle/Perception
ἅπτομαι (+GEN) touch
ἐφάπτομαι
partake of food, taste

As with other cognitive events, perceptions also differ in regard to volitionality (Kemmer
1993, 136-37; Allan 2003, 97). This semantic difference often motivates a difference in formal
coding.146 Middle verbs of perception tend to involve controlled actions; an experiencer acts
intentionally: ακροάομαι ‘listen to’; θεάομαι ‘look at’. Active verbs of perception, on the other
hand, often involve non-volitional or unintentional perceptions: βλέπω ‘see, be able to see’
(secondary sense is intentional: ‘look at, note’); κλύω ‘hear’ (secondary sense: ‘attend to’);
λεύσσω ‘see’ (secondary sense: ‘look at’). Such scalar semantic patterns among verbs, with no
definite demarcation between volitional vs. non-volitional acts, along with the natural polysemy
that arises with lexical semantics, may lead to some degree of morphological variation among
perceptions. Active and middle forms may exist as synchronic variants of the same lexeme.
(4.52) Media tantum – Perception
αἰσθάνομαι
notice, perceive, understand
δέρκομαι
see, see clearly
ἐνωτίζομαι
listen carefully to
(ἐπ)ακροάομαι listen
145

Formal variation – Perception
ἀκούω/ομαι
hear
εἰσακούω/ομαι listen, obey
ὁράω/ομαι
see, catch sight, look
παρατηρέω/ομαι observe, watch

Also: ἀπογεύω ‘give a taste of’ vs. ἀπογεύομαι ‘take a taste of’.
Most perception verbs in the middle domain receive a sigmatic aorist, reflecting their more volitional
involvement, but some are expressed with -(θ)η- as in ὀσφραίνομαι ‘smell’ (ὠσφράνθη) which is typically nonvolitional. This also applies to δέρκομαι in Plutarch, Adolescens 4: ‘As especially lucky are those among mortals
who, (οἳ ταῦτα δερχθέντες) having seen these [mysteries], will pass to Hades’. Its semantic shift in Classical from
Homeric ‘fix one’s eyes, look at’ to the more generic ‘see’ may provide sementic motivation for why δέρκομαι
adopted -(θ)η- when other perception middles had not yet done so. Its use with the sense ‘see’ suggests less
volitionality/control on the part of the experiencer, placing it closer to the patient-like end of the spectrum in contrast
to other perceptions that involve a more agent-like experiencer (θεάομαι ‘look at’), which utilize -σα- middle forms
in Hellenistic. In this way, δερχθέντες reflects a more patient-like involvement of the primary figure, analogous to
mental process middles that frequently take the -(θ)η- form in Classical and Hellenistic (Allan 2003, 159).
146

119

ἐπισκέπτομαι look at, inspect147
σκέπτομαι
look at, consider

περιβλέπω/ομαι look at, around
προοράω/ομαι see beforehand

Due to the symmetrical nature of experiencer-stimulus relations, some perceptions may
also allow variable argument realization in the clause. This, of course, depends on the lexical
semantics of the verb. If a verb lexicalizes the cognitive experience of attending to a stimulus,
such as listening or looking, then the experiencer is likely to be realized as primary figure with
the stimulus as a secondary participant (or left unrealized in a one-participant construal). The
verb profiles the relation of the experiencer directing attention to the stimulus (Croft 2012, 234).
If the verb does not lexicalize a particular direction of force, then either experiencer or
stimulus may be coded as primary figure and starting point for the action. With ἅπτομαι ‘touch’
in (4.53), the experiencer-subject serves as energy source, touching the menstruating woman in
(a), the corpse in (b). The stimulus then prompts a change in the experiencer, as ultimate energy
endpoint for the action, highlighting the altered state of the experiencer from clean to unclean.
(4.53) (a) πᾶς ὁ ἁπτόμενος αὐτῆς ἀκάθαρτος ἔσται ἕως ἑσπέρας
anyone who touches her will be unclean until evening (LXX Leviticus 15:19)
(b) Ὁ ἁπτόμενος τοῦ τεθνηκότος πάσης ψυχῆς ἀνθρώπου ἀκάθαρτος ἔσται ἑπτὰ ἡμέρας
The one who touches the corpse of any human being will be unclean for seven days
(LXX Numbers 19:11)
In (4.54) the stimulus-subject serves as starting point for the action. The profiled relation
in (a) highlights the energy transfer from the stimulus (sword) to the experiencer (people, even to
their soul). A similar interaction occurs in (b); Jesus transfers energy through his touch to a
woman’s hand to make her clean. With ἅπτομαι either experiencer or stimulus may be realized as
primary figure in a symmetrical interaction. Each formal coding provides a different construal
for the same type of event. The experiencer-subject realization is connected to a greater degree of

147

The sense ‘look at, inspect’ extends to ‘go see, visit’ and ‘look after, be concerned about’.

120

control on the part of the experiencer. In turn, the stimulus-subject realization is linked to a lesser
degree of control for the experiencer (Croft 1993, 65–67; 2012, 235). The participant chosen as
primary figure (coded as subject) is often the more salient participant, as discourse topic.
Variable realization also underscores an important point concerning the use of ‘subjectaffectedness’ as a short-hand in describing middle events. In some cases, based on the nature of
the event development associated with a particular verb, the affected experiencer may or may not
be profiled as the subject in a given middle event.
(4.54) (a) καὶ ἰδοὺ ἥψατο ἡ μάχαιρα ἕως τῆς ψυχῆς αὐτῶν
And look, the sword has touched all the way to their soul (LXX Jeremiah 4:10)
(b) καὶ ἥψατο τῆς χειρὸς αὐτῆς, καὶ ἀφῆκεν αὐτὴν ὁ πυρετός
And he touched her hand, and the fever left her (Matthew 8:15)
Cognitive events in the middle domain in §4.4 deviate from the transitive prototype in
two central ways: energy transfer and scope of attention. Because mental events rely on an
experiencer-stimulus interaction, they involve a bidirectional relational arc, with two participants
and two relations. An experiencer, as a sentient participant provides mental or perceptual
attention. In turn, the stimulus, in the virtue of the interaction, brings about a change in the
experiencer. This stands in contrast to a basic agent-patient relation, which profiles an
asymmetric energy transfer, with a unidirectional arc from agent to patient.
In the cognitive domain, middle marking is used for events in which an experiencer
directs attention toward a stimulus and subsequently experiences a change in reaction to that
stimulus. Conflating the roles of energy source and endpoint onto one participant (rather than
two distinct participants) narrows the scope of attention, highlighting the change undergone by
the experiencer. This may be cognitive, perceptual, or an expressive speech act.

121

4.5

Alternations in voice: Symmetrical energy in cyclic events

In the final subset of middle event types, alternations express a contrast in event termination, or
how an event comes to an end. In a typical agent-patient interaction, an agent (source) applies
asymmetrical force toward a patient (endpoint). In contrast, cyclic events in the middle offer an
alternate energy endpoint. Rather than terminating in a distinct patient, the effects of a cyclic
action accrue to the agent itself. The affected agent, as primary figure, plays both source and
endpoint roles in the same event (Næss 2007, 82).
Symmetrical force is present in reciprocals and reflexives. Reciprocal relations, e.g. They
met in the stairwell, are semantically collective in nature. Two or more participants are involved
in an inverse relationship. The first acts on the second; the second acts on the first in the same
way. In a typical reflexive, an agent acts on himself. This may be direct, e.g. He shaved, with an
agent acting on his own body. Or, it may be indirect, e.g. She adopted a son, in which case an
agent acts on a distinct patient but is affected by the outcome as a recipient or beneficiary
(Kemmer 1993, Shibatani 2006, 232; Croft 2012, 236).
Their commonality comes in the bidirectional transmission of force that results in dual
roles for the primary figure. Among these more agent-like middle types, the primary figure is a
volitional agent, initiating the event – whether reciprocal or reflexive. Thus, the primary figure
plays dual source and endpoint roles as an affected agent. Where they differ is in the directness
of the agent’s involvement. For typical reciprocal relations, the agent is a direct source and
endpoint, but since there are multiple participants, all such figures play the same dual roles. For
reflexives, the agent is a direct source and endpoint of his own bodily action. But for indirect
reflexives, the agent, as a direct source, is only an indirect endpoint – as recipient or beneficiary
of an action carried out on a secondary figure (Croft et al. 1987, 186).

122

In this regard, cyclic events deviate from the asymmetric transmission of force in the
prototypical transitive, but they do so differently from middle events in the cognitive domain
(§4.4). Among cognition middles, there are two directions of force, each of a different event
type. In the first, an experiencer pays attention to a stimulus. In the second, a stimulus induces a
change in an experiencer. For cyclic events, the two directions of force are of the same type.
With reciprocals, two participants act on each other in the same way, e.g. embrace. With
reflexives, there is just one cyclic action, and thus only one event type, e.g. wash oneself
(Kemmer 1993, 54-81, 202; Shibatani 2006; Croft et al. 1987; Croft 2012, 235-37).
4.5.1

Reciprocal

Reciprocal relations, like events in the cognitive domain, rely on a symmetrical transmission of
force, with two participants and two relations. But rather than representing different event types
(as in experiencer-stimulus interactions), the two directions of force for reciprocals involve
inverse relations of the same type. Participants act on one another in similar ways. And because
of their relational symmetry, they play identical roles: Each participant is energy source of one
relation and energy endpoint of a second relation of the same type.
This kind of symmetrical energy deviates from the asymmetrical force associated with
the transitive prototype. In a typical agent-patient interaction, two distinct participants play two
distinct roles. The process highlights the agent’s activity vis-à-vis its interactions with a second
figure. Beginning with an energy source, the event follows the effects of the action as it plays out
on a distinct endpoint. For reciprocals, this participant distinction is lost, as participants play
essentially the same roles. Figure 8 illustrates this role conflation in reciprocal events, which
include verbs that naturally involve mutual exchange, e.g. meet, fight, argue (Kemmer 1993,
102; Næss 2007, 23; Nedjalkov 2007c, 6; Croft 2012, 236).

123

Figure 8

Reciprocal event

This lack of distinction has the effect of narrowing the scope of attention. For reciprocals,
participants act collectively in the same event at the same time. If A acts on B and B acts on A,
then A and B act jointly. Removing the conceptual distinction between them draws attention to
the joint character of their action. As a consequence, onstage focus does not shift from primary
source to secondary endpoint. Instead, attentional focus is drawn to the relational symmetry of
simultaneous source/endpoint figures. The joint action links two participants to each other
without shifting to a distinct endpoint (Kemmer 1993, 101; Næss 2007, 29; Croft 2012, 244).
Reciprocals permit either one- or two-argument structures. Each formal coding represents
a different semantic construal. For one-argument structures, two symmetrically interacting
participants are construed in terms of a plural set, as a holistic entity (Nedjalkov 2007c, 8).
(4.55) Reciprocal Middle
μάχωνται δύο ἄνδρες
Two men are fighting (LXX Exodus 21:22)
καὶ εὗρεν αὐτοὺς περιπλακομένους καὶ κλαίοντας
She found them embracing and crying (Testament of Abraham (A) 5.11)
As a group, participants act collectively in the same dual roles. Source and endpoint are fused
into a cluster of symmetrical relations. This allows two simultaneous sub-events (A acts on B, B
acts on A) to be conceived as a collective one-participant action. Like collective motion middles
(§4.3.3), reciprocals are internally caused. The energy source is internal to the group itself,
allowing for a one-argument construal (Allan 2003, 85-86; Croft 2012, 236-245).
In one-argument reciprocals like (4.55), participants form an undifferentiated whole; no
one participant is given primary attention over another. Instead, the collective nature of their

124

simultaneous action is highlighted. Two-argument reciprocals offer an alternate construal. In a
two-argument conception, participants are differentiated into two referential entities.
(4.56) διελέγετο μὲν οὖν ἐν τῇ συναγωγῇ τοῖς Ἰουδαίοις
So he argued in the synagogue with the Jews (Acts 17:17)
(4.57) καὶ ἐμαχέσαντο οἱ ποιμένες Γεραρων μετὰ τῶν ποιμένων Ισαακ
And the shepherds of Gerares fought with the shepherds of Isaac (LXX Genesis 26:20)
One (of two) participants is chosen as primary figure (usually discourse topic), expressed
as nominative subject, with the other as a secondary figure, expressed as dative complement in
(4.56), or prepositional phrase in (4.57). In (4.56), the outgoing energy from Paul to the Jews is
highlighted as focus of attention. The incoming energy from the Jews to Paul is present but
comparatively backgrounded. Here, Paul’s perspective comes into focus. His distress at finding
idols in Athens is what brings him to argue back and forth with the Jews It is the symmetrical
energy in reciprocals that enables either participant to be chosen as focal point (Allan 2003, 8687). In (4.58) both realizations take effect with the inclusion of a subordinate clause.
(4.58) ἄνδρες γοῦν οὐ γυναιξὶν οὐδὲ γυναῖκες ἀνδράσιν ἁμιλλήσαιντο
Men could not contest with women, nor women with men (Philo Sacrifices 100)
While reciprocals represent cyclic action, the constraints of argument structure in a basic
clause require an acyclic realization of arguments. Consequently, reciprocals are expressed much
like acyclic events, realized as either two-argument structures with one a primary and secondary
figure, or as one-argument expressions with a holistic construal. The cyclic nature of such events
is signaled grammatically, via middle morphology in middle-marking languages. In (4.57),
argument structure encodes one direction of force; middle morphology signals that the reverse

125

relation is also true. Middle marking explicitly encodes that the energy source is also the energy
endpoint of a symmetrical relation (Allan 2003, 86; Nedjalkov 2007c, 8-11; Croft 2012, 240).148
In addition to this, two-argument reciprocals in Greek are often realized with a dative
complement (or prepositional phrase) rather than the default accusative in the prototypical
transitive (Kemmer 1993, 107). This shift in formal coding signals a difference in construal in
how participants are involved in the action. A prototypical patient (coded as accusative in Greek)
undergoes the effects of an action without any kind of initiating role in the event. Yet the
comitative role, signaled by the use of the dative among two-argument reciprocals, signifies
collaborative behavior in a joint activity. The secondary figure in a reciprocal is involved in the
same type of action in a similar way as the primary figure. In English, this role is often marked
by a prepositional phrase, e.g. The man fought with/against his attacker (Croft 2012, 245). This
is similar to μάχομαι ‘fight’. It may be realized with a dative complement, e.g. μαχόμενος τῷ
νόμῳ κυρίου ‘fighting with the law of the Lord’ (Protevangelium of James 14:1). Or, the
preposition ἐπί offers an alternative construal with a focus on the hostile opposition between
parties: ἐπʼ Αὐσονίοισι μάχεσθαι ‘to fight against the Ausonians’ (Sibylline Oracles 13.141).
Voice alternations among reciprocals express a contrast in symmetrical vs. asymmetrical
energy. Note the use of δια- with some middles, reflecting a sense of separation and mutual
exchange. In such cases, the active encodes asymmetrical force in what is usually a oneparticipant action (though plural subjects occur, e.g. ‘they spoke to him’). The derived middle

148
Cyclic events may also be lexically expressed as a facet of lexical semantics. A few reciprocal verbs are
realized with active morphology rather than middle. In such cases, cyclic action is signaled lexically based on the
meaning of the verb rather than signaled grammatically: active πολεμέω ‘fight, do battle with’ (πολεμοῦντα τοῖς
Ἰουδαίοις ‘at war with the Jews’ (Josephus Wars of the Jews 7.423)); active ἐρίζω ‘quarrel, vie with’ (ἐρίζουσαι τῷ
Ισαακ ‘quarreling with Isaac’ (LXX Genesis 26:35)).

126

with δια- denotes symmetrical energy among two or more participants in a synergistic relation
that results in combat, competition, or conversation (Allan 2003, 86; Nedjalkov 2007c, 95-96).
(4.59) Active/Asymmetrical
ἀμείβω
change, exchange
καταλλάσσω change, reconcile
μίγνυμι
bring together, mix
περιπλέκω twine around, bind
ἀκοντίζω
λέγω
τοξεύω

hurl a javelin
speak
shoot with a bow

Middle/Symmetrical
ἀμείβομαι
take turns, alternate, repay
καταλλάσσομαι
become reconciled with
μίγνυμαι
have sexual relations with
περιπλέκομαι
embrace
διακοντίζομαι
διαλέγομαι
διατοξεύομαι

compete/combat with javelins
converse, discuss, argue
compete in archery

(4.60) Media Tantum – Symmetrical
ἀγωνίζομαι
fight, struggle149
μάρναμαι
ἀκροβολίζομαι skirmish
μάχομαι
ἁμιλλάομαι
compete, vie, contend with συγγίγνομαι

contend
fight, quarrel150
have sex with

Middle morphology in Greek and other middle systems is used for naturally reciprocal
actions, those events that typically involve two or more participants acting on one another. The
reciprocal pronoun ἀλλήλους is then used among non-reciprocal events to signal a reciprocal
interpretation with symmetrical energy rather asymmetrical force. One of the key differences
between middle morphology and reciprocal pronouns is in their distribution. The middle form
occurs with verbs whose lexical meaning necessarily or at least frequently implies a mutual
exchange between participants. Reciprocal pronouns, in turn, are used with a wider set of verbs,
any event that is not naturally reciprocal (Kemmer 1993, 102-3; Allan 2003, 84-86).
Which verbs represent reciprocal action, and which do not, is grounded in human
experience – pertaining to events that involve mutual human relations. At the same time, it is
also language specific, arising within the constraints of a socio-cultural context. English kiss is
frequently reciprocal though not necessarily so. They kissed is typically interpreted as a

149
150

Also: ἐπαγωνίζομαι ‘contend again’; συναγωνίζομαι ‘fight along with’ focuses on the supportive role.
Also: διαμαχομαι ‘fight against each other, do battle with, argue sharply’.

127

simultaneous exchange between two participants, often in the context of lovers: A kissed B, B
kissed A.151 In Greek, καταφιλέω ‘kiss’ is typically interpreted as an asymmetrical action with
active morphology rather than middle. It is often used in contexts of greeting and leave taking.
To express a reciprocal action, a reciprocal pronoun is used (Kemmer 1993, 105).
(4.61) κατεφίλουν ἀλλήλους, περιπλεκόμενοι τοῖς συγγενέσιν, ἐπὶ τοὺς τραχήλους
they began to kiss one another, embracing their kinsmen on their necks (3 Macc. 5:49)
This illustrates a natural semantic constraint in middle marking. The middle form occurs
with events that are naturally reciprocal, a grammatical expression of a semantic category, i.e.
those events conceived as typically reciprocal in nature. The same occurs with direct reflexive
middles. Middle marking is used to express typical grooming actions and body care events.
4.5.2

Grooming/Direct reflexive

Reflexive and reciprocal relations are similar in kind. In both, participant involvement relies on
the same dual roles – source and endpoint – of a cyclic action. The subject in each is an animate
and agentive figure who performs the same action that it undergoes. The primary figure is thus
the source and endpoint of a symmetrical transmission of force. The difference between the two
comes in both the nature and number of participants involved. For reciprocals, two separate
participants act, not on themselves, but on one another. For direct reflexives, this action is turned
inward. A single primary figure acts on himself, rather than another participant.
Middle marking is used with grooming actions – those typically or naturally performed
by humans on their own bodies. The primary figure is both source as initiator of the action and
endpoint as the one changed by it. Unlike the transitive prototype, source and endpoint roles are
151

English has just one marker of reciprocal semantics. It occurs with transitive verbs but is not required
with naturally reciprocal actions. These are expressed as unmarked intransitive events. Reciprocal semantics may be
expressed by omitting the direct object and expressing the intransitive with a plural subject. This strategy is
constrained to those events that are conceived as naturally reciprocal: She kissed him > They kissed; She met him >
They met; She argued with him > They argued (Kemmer 1993, 102; Nedjalkov 2007a, 171).

128

conflated onto a single referent, as in Figure 9. Onstage focus is drawn toward the primary figure
who is changed by his own action (Langacker 1991, 367; Kemmer 1993, 98; Croft 2012, 245).

Figure 9

Direct reflexive/Grooming

Reflexives, like reciprocals, permit one- and two-argument structures; each reflects a
different construal. In one-argument conceptions, two semantic roles converge onto a single
entity. And because such actions are frequently carried out by people on their own bodies, no
object coding is required (Kemmer 1993, 60). It is often the case that the lexical semantics of the
verb and the context of use constrain the interpretation, helping to distinguish which body part is
affected. Verbs like περικλύζομαι ‘wash one’s body’ and νίπτομαι ‘wash a body part’ (e.g. hands,
feet, head) may illustrate. In (4.62), a man goes down to the Tigris River to wash his body. In
(4.63), the context of entering the tabernacle, along with the use of the verb νίπτομαι, rather than
περικλύζομαι, implies that this particular act of washing refers to the hands and feet (not the
whole body), as instructed by God for the purification of priests.
(4.62) τὸ δὲ παιδάριον κατέβη περικλύσασθαι
Then the young man went down [to the river] to wash (LXX Tobit 6:3)
(4.63) νίψονται ὕδατι καὶ οὐ μὴ ἀποθάνωσιν
they will wash with water, so they will not die (LXX Exodus 30:20)
With regular body actions, middle morphology is used to signal their semantic character:
source and endpoint roles are naturally fused together onto a single participant who acts on
himself. For περικλύζομαι and νίπτομαι, the endpoint participant is not passively involved as in a
typical agent-patient interaction. Instead, the endpoint-self actually participates in the event in

129

some way. By necessity, acts of washing (along with other grooming events) require the use of
the body and body parts as simultaneous acting and acted on entities. With the self-involved
nature of such actions, it makes sense that typical body care verbs would formally reflect, via
middle morphology, a conceptual difference from actions that are normally carried out on a
distinct endpoint participant (Kemmer 1993, 60).
What accounts for their single-argument conception is the internal causality associated
with such events. The energy that brings about the action is internal to the single participant, as
source and endpoint of the same event. This is also what brings the grooming/reflexive middle
into close proximity with other one-participant middle types, such as anticausative alternations,
expressing a change of state among physical processes (§4.3.1), body motion (§4.3.2), collective
motion (§4.3.3), and mental processes (§4.3.4). For both anticausative and reflexive middles, the
energy source is internal to the changed participant, rather than coming from an external source
of energy. In each case, the primary figure, as the focus of attention, is the changed participant –
the one that undergoes the action of the verb.
The difference among anticausatives and direct reflexives is the degree of control on the
part of the primary figure. Both grooming and body motion middles involve bodily action,
wherein a more agentive figure volitionally instigates the action that it also undergoes. In
comparison, other one-participant middles (e.g. physical processes, mental processes, and
collective motion), tend to involve a lower degree of control on the part of the primary figure.
The single participant is more patient-like, with far less control over bringing about the action.
This suggests that the grooming middle shows more of a difference in degree than in kind with
other one-participant middle events. On a scale of one-participant actions, grooming verbs

130

occupy the higher-control end of the continuum, with physical processes at the lower-control end,
showing little to no control over the process described by the verb (Croft 2012, 238).
(4.64) Degree of control: One-participant actions
Physical process
ἀπόλλυμαι
die
Mental process
ἐκλανθάνομαι forget
Collective motion διαλύομαι
disperse
Body motion
ἐγείρομαι
stand up
Grooming
ἀμφιέννυμαι
dress

lower

higher

Beyond one-argument conceptions, there are also two-argument construals. In twoargument reflexives, a second argument is overtly realized in the clause that is coreferent with
the first. Middle morphology signals an identity of relational participants. For most grooming
events, this is a part-whole possessive relation. A possessor (agent) stands in possession of a
partitive (patient). Supplying this secondary object not only provides an explicit specification,
identifying the locus of effect upon which the agent acts, but also a conceptual separation of
different facets of a single referent, i.e. a volitional agent and the affected body part.
The lexical semantics of a given verb and the grooming action it denotes naturally
constrain the noun phrase which may be expressed in this secondary position. Bodily actions
directed toward the self involve inalienable or at least quasi-inalienable possession. Verbs like
περικλύζομαι ‘wash one’s body’ and χρίομαι ‘anoint oneself’ in (a) typically refer to whole-body
actions without specifying a particular body part. But κατακαλύπτομαι ‘wear a veil, cover one’s
face/head’ in (b) and περιτέμνομαι ‘circumcise oneself’ in (c) generally refer to body actions
directed toward a specific body part rather than the whole (Langacker 1991, 368; Kemmer 1993,
77; Genuišienė 1987, 131; Dench 2013, 145).
(4.65) (a) περιεκλύσατο τὸ σῶμα ὕδατι καὶ ἐχρίσατο μύρῳ παχεῖ
she washed her body with water, and anointed [it] with rich perfume (LXX Judith 10:3)
(b) Ιουδας ἔδοξεν αὐτὴν πόρνην εἶναι, κατεκαλύψατο γὰρ τὸ πρόσωπον αὐτῆς
Judah assumed she was a prostitute, for she covered her face (LXX Genesis 38:15)

131

(c) περιετέμετο τὴν σάρκα τῆς ἀκροβυστίας αὐτοῦ
he circumcised the flesh of his foreskin (LXX Judith 14:10)
With verbs of dressing and undressing, reflexive actions typically involve articles of clothing and
other wearable items. The second argument specifies which garment is put on or taken off.
(4.66) ἐξεδύσατο τὰ ἱμάτια τῆς χηρεύσεως αὐτῆς καὶ ἐνεδύσατο τὰ ἱμάτια τῆς εὐφροσύνης αὐτῆς
she took off her garment of widowhood and put on her garment of joy (LXX Judith 10:3)
A lexeme like ζώννυμαι ‘gird oneself’ is used not only for girding one’s clothes with a belt for
work, but it is also in contexts of battle, preparing to fight by strapping on one’s sword or armor.
(4.67) Ζώσασθε ἕκαστος τὴν ῥομφαίαν
Everyone strap on his sword (LXX 1 Kingdoms 25:13)
The middle, unlike the prototypical transitive, lacks an expectation of another participant. The
agent acts on a patient, but rather than a distinct endpoint, the patient represents part of the
agent’s own body or something that covers the agent’s body, as a metonymic relation. The
normal expectation among such grooming events is the identity of source/endpoint roles.
The consequence of this expectation is that middle morphology is not likely to occur on
any generically self-directed action, but this event type requires socio-culturally recognized
activities that are normally performed on oneself as a regular practice. Middle formations in
(4.68) do not just indicate that the lexical meaning of the verb (cutting, covering) is directed
toward oneself in a general way, but they specify culturally relevant acts performed on the body.
That is, ἀποκόπτομαι (in opposition to ἀποκόπτω) does not mean ‘cut oneself’ but refers to
making oneself a eunuch, an act that specifies a culturally sanctioned cutting of a specific body
part. The same may be said of κατακαλύπτομαι. It is not used as a generic act of covering oneself
but refers to the use of a veil to cover one’s head or face in a culturally appropriate manner.
(4.68) ἀποκόπτω
cut away s.t.
κόπτω
cut off s.t.
κατακαλύπτω cover s.t./s.o.

ἀποκόπτομαι
κόπτομαι
κατακαλύπτομαι

castrate oneself
smite or beat oneself (in mourning)
put on a veil, cover one’s head

132

Regular grooming activities may involve culturally-specific events like rites of purification or
circumcision, but they also include more cross-cultural activities like dressing and keeping clean.
(4.69) Active/Causative
ἁγνίζω
make clean, purify
βαπτίζω
make clean, purify
θερμαίνω
make warm, heat
καθαρίζω
make clean, purify
ὁπλίζω
make ready
παρασκευάζω make ready
παύω
cause to stop
(4.70) Active/Acyclic
Ritual activities
ἀλείφω
anoint with oil
ἀποκόπτω
cut away
κόπτω
cut off
περιρραίνω
sprinkle around
περιτέμνω
circumcise s.o.
στεφανόω
crown/wreath s.o.
χρίω
rub, anoint with oil
Washing
(ἀπο)λούω
wash (off)
ἀπομάσσω
wipe off, clean
(ἀπο)νίπτω
wash (off)
περικλύζω
wash all around
Taking off
ἀποβάλλω
throw off
ἐκδύω
strip off
ἐκτινάσσω
shake off, expel
περιαιρέω
strip off, remove
Putting on
ἀμφιέννυμι
clothe, dress s.o.
ἐνδιδύσκω
clothe s.o.
ἐνδύω dress clothe s.o.
ἐπισκευάζω equip, repair
κατακαλύπτω cover s.t./s.o.
περιβάλλω
encompass, clothe
περιδέω
bind, wrap around
152

Middle/Anticausative
ἁγνίζομαι
purify/dedicate oneself
βαπτίζομαι
get baptized, purify/wash oneself
θερμαίνομαι
warm oneself, get warm
καθαρίζομαι
purify oneself
ὁπλίζομαι
get ready, prepare oneself
παρασκευάζομαι get ready, prepare oneself
παύομαι
cease an action, stop oneself152
Middle/Cyclic
ἀλείφομαι
ἀποκόπτομαι
κόπτομαι
περιρραίνομαι
περιτέμνομαι
στεφανούμαι
χρίομαι

anoint oneself
castrate oneself
beat oneself (in mourning)
sprinkle oneself in purification
circumcise oneself, get circumcised
crown oneself
anoint oneself

(ἀπο)λούμαι
ἀπομάσσομαι
(ἀπο)νίπτομαι
περικλύζομαι

bathe, wash (off) oneself
wipe off oneself
wash (off) part of the body
bathe all around

ἀποβάλλομαι
ἐκδύομαι
ἐκτινάσσομαι
περιαιρέομαι

throw away from oneself
strip off oneself153
shake off oneself
take off oneself

ἀμφιέννυμαι
ἐνδιδύσκομαι
ἐνδύομαι
ἐπισκευάζομαι
κατακαλύπτομαι
περιβάλλομαι
περιδέομαι

dress, get dressed
dress, wear
wear, put on154
get ready, equip oneself
wear a veil, cover one’s head
clothe oneself, put on
wrap around oneself

Also: ἀναπαύω ‘cause to rest’ vs. ἀναπαύομαι ‘take one’s rest, rest, cease an activity’.
Also: ἀποδύω ‘strip off’ vs. ἀποδύομαι ‘strip off oneself’.
154
Also: ἐπενδύ(ν)ω ‘put on over’ vs. ἐπενδύομαι ‘put on, wear an outer garment’.
153

133

(περι)ζώννυμι gird about
ὑποδέω
fasten under, bind
Cutting hair
κείρω
sheer, cut
ξυράω
shave s.o.
Violent activities
(ἀπ)άγχω
strangle, choke s.o.
σπάω
drag, pull, draw
(4.71) Media tantum – Cyclic
ἀπεκδύομαι strip off oneself
ἐγκομβόομαι wear, put on

(περι)ζώννυμαι gird oneself, put on one’s belt155
ὑποδέομαι
strap on one’s shoes
κείρομαι
ξυράομαι

get a haircut, have one’s hair cut156
shave, get shaved (e.g. head/beard)

(ἀπ)άγχομαι
σπάομαι

hang oneself, strangle oneself157
draw one’s sword

ἐγκρατεύομαι control oneself, abstain
περίκειμαι
put around oneself, wear158

The middle form, as noted in §4.5.1, is semantically constrained. It is used to mark
regular grooming actions, those that typically involve an agent acting on himself. Source and
endpoint roles are expected to be filled by the same referent. The reflexive pronoun ἑαυτὸν is
then used among acyclic actions to signal a reflexive interpretation with symmetrical energy
rather than the default asymmetrical force. The pronominal form occurs with a wider set of
verbs, any events that are not typically reflexive in nature (Kemmer 1993, 71; Allan 2003, 90).159
In this way, suggesting that the two forms are mere alternatives of one another with the
same semantic content ignores differences in their distribution. Such patterns are substantiated
both cross-linguistically among middle systems and intralinguistically in Greek. Even among
verbs of washing there are different semantic expectations that lead to differences in formal

155
Also: ἀναζώννυμι ‘gird up again’ vs. ἀναζώννυμαι ‘gird oneself for work/battle’; διαζώννυμι ‘gird
round, encircle’ vs. διαζώννυμαι ‘tie around oneself’; συζώννυμι ‘gird s.o.’ vs. συζώννυμαι ‘gird oneself’.
156
Also: περικείρω ‘cut around’ vs. περικείρομαι ‘clip one’s hair’.
157
Actions like hanging oneself or stranging oneself, note that these involve ways that a person would be
able to kill themselves – actions that are unlikely to be performed on the self do not occur among direct reflexive
middles – these include torture, crucifixion, etc.
158
From κεῖμαι ‘be in a place, lie’: περί + κειμαι ‘be positioned around’ > ‘put s.t. around, wear s.t.’
159
One of the functions of the reflexive pronoun ἑαυτὸν is to signal contrastive stress, denoting that the
object explicitly refers to the same referent as the subject and not some other potential referent in the discourse
(Kemmer 1993, 63). In doing so, the reflexive pronoun can and does occur with middle-marked grooming verbs. In
such cases, source/endpoint roles are already expected to be filled by the same referent. The addition of the reflexive
pronoun signals the speaker’s emphasis on this fact, explicitly marking the identity of relations, especially in cases
where there may be some ambiguity about other potiential referents in the discourse context.

134

expression. A verb like ἐκκαθαίρω ‘cleanse’ is most often used for purging a house or other
vessel of unclean elements. It represents a ritual cleansing directed at a distinct endpoint figure
and is not regularly used in relation to one’s own body. In (a), the active form occurs with a
reflexive pronoun, to signal an unexpected coreference of participants, based on an analogical
extension: A person’s body is like a house or vessel that can be purged of unclean things.
(4.72) (a) ἐὰν τις ἐκκαθάρῃ ἑαυτὸν ἀπὸ τούτων, ἔσται σκεῦος εἰς τιμήν
If someone cleanses himself of these, he will be a vessel of honor (2 Timothy 2:21)
(b) ἐὰν μὴ πυγμῇ νίψωνται τὰς χεῖρας οὐκ ἐσθίουσιν
they do not eat unless they thoroughly wash their hands (Mark 7:3)
Example (a), marked with a reflexive pronoun, represents a case in which a normally acyclic
event is marked as a cyclic action. On the other hand, a ritual washing in (b) that is regularly
directed toward one’s own body is expressed with middle morphology, signaling the expected
identity of source and endpoint roles. The Pharisees and Jews wash their hands before eating.
4.5.3

Self-benefactive/Indirect reflexive

Direct and indirect reflexives share a similar relational structure. Both depict self-oriented
construals: A primary figure instigates an event, as the head of an action chain, and is also
affected by it, as the tail of a relational arc. The primary figure is thus an affected agent, a
participant who volitionally instigates an action but is also construed as being affected in some
way by the outcome of the event (Næss 2007, 82). For direct reflexives, an agent acts on his own
body. If a second argument is realized in the clause, it is coreferential with the first.160 For
indirect reflexives, an agent acts on a secondary figure but does so as a recipient or beneficiary of
the event. Unlike direct reflexives, the indirect type involves a distinct patient that is acted on by
the primary figure, hence its two-argument realization in the clause (Croft 2012, 240).

160

Or: something that covers the body, or a possession worn on the body.

135

Despite its two-argument construal, the indirect reflexive still supplies a contrast in event
termination, or how an event comes to an end. In a typical agent-patient relation, asymmetrical
force terminates in a final endpoint that is distinct from the source. Acyclic action is not the only
way to construe event development. Symmetrical force in indirect reflexives yields an alternate
endpoint. Rather than terminating in a distinct patient, the action accrues back to the agent, as
source and indirect endpoint of the same event. The spotlight is drawn to the primary figure who
is affected by his own action (Kemmer 1993, 78; Langacker 2006, 130; Croft 2012, 236).
Alternate construals in event development lead to cross-linguistic differences in formal
expression. In middle systems, conflation of source and endpoint roles is signaled grammatically
via middle morphology. Argument realization encodes one direction of force (source to
endpoint). Middle morphology encodes the reverse relation (endpoint to source), explicitly
marking the cyclic nature of the action, as in Figure 10 (Croft 2012, 235-40).

Figure 10

Indirect reflexive

Lexical semantics and event development play a major role in the distribution of the
middle form. Middle morphology is semantically constrained to events that are typically or
necessarily performed with the agent as a recipient or beneficiary of the action. That is, the
conflation of participant roles – agent (source) and recipient/beneficiary (indirect endpoint) – is
an expected part of the meaning of the verb. This includes events in which the primary figure is a
recipient (4.73) as well as a beneficiary (4.74). In both cases, the middle form is used for events
in which the primary figure is the intended endpoint.

136

For middle-only verbs in (4.73), the primary figure is a natural recipient of the action.
The subject receives or brings s.o./s.t. toward himself, highlighting the energy transfer from
endpoint to source. This may apply to physical relations (take into one’s arms) or to
possession/acceptance (receive, acquire, use) (Kemmer 1993, 78-81).
(4.73) δέχομαι
accept, receive, take161 ἐπιλαμβάνομαι take hold of, grasp, catch
δράσσομαι
grasp by the hand, seize κτάομαι
acquire, get, possess
ἐναγκαλίζομαι take in one’s arms
χράομαι
use, treat, deal with162
In (4.74), the middle form denotes self-benefactives – events in which a typical part of
the meaning of the verb is that the effects of the action accrue to the agent as a beneficiary. In
Greek, this includes commerce and business – buy, trade, work, earn – as well as trickery and
deception, allowing the primary figure to take advantage of the secondary participant.
(4.74) διαπραγμᾰτεύομαι
ἐργάζομαι
πραγματεύομαι
ὠνέομαι
κατασοφίζομαι
παραλογίζομαι
ὑποκρίνομαι

gain by trading, earn
work, earn, trade163
busy oneself, do business, trade
buy
take advantage of by trickery
defraud s.o., take advantage of, mislead
pretend, act hypocritically, play a part

For events in (4.73) and (4.74), there is an expectation of participant coreference, in
which source and endpoint roles are conflated onto the primary figure, as an affected agent. If
such events constitute the semantic core of the indirect reflexive type, then events in (4.75) may
be considered extensions of that core, including verbs of pillaging, disarming, and overpowering
(often in battle) as well as verbs of giving, healing, and saving. In (a), the primary figure gains
advantage/authority over his foe, by plundering, attacking, or conquering the enemy. For some,

161

Derivatives: ἀναδέχομαι, ἀπεκδέχομαι, ἀποδέχομαι, διαδέχομαι, εἰσδέχομαι, ἐκδέχομαι, ἐπιδέχομαι,
παραδέχομαι, προσδέχομαι, ὑποδέχομαι.
162
Derivative: καταχράομαι ‘use up, make full use of, misuse’.
163
Derivatives: ἀπεργάζομαι, κατεργάζομαι, περιεργάζομαι, προσεργάζομαι.

137

the primary figure is also a recipient, as with ἀπεκδύομαι when he disarms his enemy, strips him
of his weapon, and gains possession of it. In (b), the primary figure is more positively disposed
toward the secondary participant. From a position of relative advantage, the subject may use his
position to show pity on the secondary figure, grant a request, show favor, or spare from loss.
(4.75) (a) Ill-disposed
αἰκίζομαι
torture, persecute
ἀπεκδύομαι
disarm164
ἐπίκειμαι
press upon, attack, urge166
ληΐζομαι
plunder, carry off booty
καταγωνίζομαι conquer, overcome
βιάζομαι
gain by force, overpower
χειρόομαι
overpower, subdue

(b) Well-disposed
ἀκέομαι
heal, cure, heal an offense
ἰάομαι
treat, heal, repair165
φῐλοφρονέομαι show favour to, embrace
δωρέομαι
grant, bestow
χαρίζομαι
show favor, give freely167
φείδομαι
spare from loss, have pity on
ῥύομαι
rescue, save from danger

Alternations in (4.76) involve a contrast in energy transfer. In the active, an acyclic action
transfers energy away from the primary figure toward a distinct endpoint participant. The middle
offers an alternative construal. Energy is transferred toward the primary figure as a recipient or
beneficiary of the action. Source and endpoint roles are conflated onto the affected agent.
(4.76) Active/Acyclic
αἱρέω
overpower, kill
ἀναβάλλω
throw s.t. up/off
ἀποδίδωμι
give up, give back, pay
δανείζω
lend
164

Middle/Cyclic
αἱρέομαι
ἀναβάλλομαι
ἀποδίδωμαι
δανείζομαι

choose, prefer168
put off for oneself, delay169
sell
borrow

Also, with a direct reflexive sense, ‘strip off clothing’.
A difference in meaning is reflected in a difference in form: sigmatic middle is agentive ‘heal’, with the
-(θ)η- form as a non-agentive, physical spontaneous process: ‘become well, recover’.
166
From κεῖμαι ‘be in a place, lie’: ἐπί + κειμαι ‘lie upon’ > attack with force > urge to do’.
167
Different lexemes occur with different semantic associations and cultural connotations. Contrast active
δίδωμι ‘give’ with middle χαρίζομαι ‘give graciously’, a verb that is traditionally treated as ‘active in meaning’
despite its middle morphology. Both verbs involve acts of giving, but it is χαρίζομαι that is used with a specific
connotation. In Classical texts, it is often used with regard to making oneself amenable or agreeable to others (e.g.
χαριζόμενος τάδʼ ἀείδει ‘he obliges with this song’ Homer, Odyssey 8.536). In Hellenistic, the verb is often used to
convey something about the nature or beneficence of the giver. The giver shows himself to be gracious or civic
minded by his act of giving. In this way, its middle expression is motivated by its verbal use, often in contexts of
ascribing honor to the giver (e.g. οὗτος ὑμῖν τούτους χαρίζεται τοὺς λόγους δι ̓ ἑρμηνέως ἐμοῦ ‘it is he (the Lord) who
bestows these instructions upon you through me as interpreter’ Josephus, Antiquities 3.87).
168
Derivatives: ἀναιρέω ‘take up, get rid’ vs. ἀναιρέομαι ‘adopt, claim’; ἀφαιρέω ‘separate’ vs. ἀφαιρέομαι
‘take away’; ἐξαιρέω ‘tear out’ vs. ἐξαιρέομαι ‘select out, save’; προαιρέω ‘produce’ vs. προαιρέομαι ‘prefer, choose’.
169
As a legal term ‘adjourn trial, delay a hearing’; also, body care sense: ‘throw over the shoulder > wear.’
165

138

ἐκδίδωμι
ἐκλέγω
ἐνδείκνυμι
ἐπισπάω
κομίζω
μερίζω
μισθόω
νέμω
νοσφίζω
ὀνίνημι
προσλαμβάνω
προχειρίζω
συμβουλεύω
τίθημι
ὠφελέω

give out, pay
point out
mark, indicate
drag, pull, draw
bring, carry off
divide, separate
let out for hire, lease out
pasture, graze a flock
set apart, separate s.t.
profit, benefit, help s.o.
add, increase
hand over (rare)
give advice, advise
lay, put/set in a place
provide benefit to

ἐκδίδωμαι
lease, rent out
ἐκλέγομαι
choose, select for oneself170
ἐνδείκνυμαι
display oneself, exhibit171
ἐπισπάομαι
draw to oneself, entice s.o.172
κομίζομαι
receive, get back
μερίζομαι
divvy up with others173
μισθούμαι
hire, employ s.o.
νέμομαι
feed, graze (of cattle)
νοσφίζομαι
hold back for oneself, steal
ὀνίνημαι
have benefit of, profit from
προσλαμβάνομαι take in/along/aside
προχειρίζομαι
choose, select
συμβουλεύομαι ask advice, consult, plot with
τίθημαι
make for oneself, establish174
ὠφελέομαι
derive profit, gain, benefit from

The middle form in (4.76) is used to denote events in which the primary figure is typically the
intended endpoint, as with νοσφίζομαι ‘put aside for oneself, pilfer, skim off’.
(4.77) καὶ ἐνοσφίσατο ἀπὸ τῆς τιμῆς, συνειδυίης καὶ τῆς γυναικός, καὶ ἐνέγκας μέρος τι παρὰ τοὺς
πόδας τῶν ἀποστόλων ἔθηκεν
He purloined some of the proceeds, and with the knowledge of his wife, he brought only
part [of the money] and laid it at the apostles’ feet (Acts 5:2)
Because of this, the middle form represents a more basic or unemphatic way of
expressing self-benefactives. The reflexive pronoun is then used with a wider set of verbs
(including those that are not typically self-benefactive) to provide contrastive stress or emphasis
on the fact that the agent and beneficiary/recipient are coreferential, i.e. that the event should be
interpreted with symmetrical energy rather than asymmetrical force, as in (4.78) with active
ἐπισωρεύω ‘pile s.t. up’ and a reflexive pronoun (Kemmer 1993, 74-81; Allan 2003, 114).

170

Additional: ἐπιλέγω ‘identify, name’ vs. ἐπιλέγομαι ‘choose, select’.
Additional: ἐπιδείκνυμι ‘display, show’ vs. ἐπιδείκνυμαι ‘show off, give prove, reveal oneself’.
172
Also extends to medical term in which the action is performed directly on a body part as a direct
reflexive: ‘draw toward oneself’ > ‘pull over the foreskin over the penis’, as in 1 Cor. 7:18.
173
Also: διαμερίζομαι ‘divvy up, divide in a group, share in’; συμμερίζομαι ‘take a share in with’
174
Derivatives: ἀποτίθημι ‘take off, lay down’ vs. ἀποτίθημαι ‘put away from oneself, avoid, stow away’;
ἀντιδιατίθημι ‘retaliate’ vs. ἀντιδιατίθημαι ‘offer resistance’.
171

139

(4.78) κατὰ τὰς ἰδίας ἐπιθυμίας ἑαυτοῖς ἐπισωρεύσουσιν διδασκάλους
they will heap up for themselves teachers according to their own desires (2 Timothy 4:3)
The use of the middle form in the cyclic domain represents a shared formal coding for a
similar relational structure. Middle morphology occurs with those events that typically or at least
frequently denote symmetrical energy, as in naturally reciprocal events (fight, embrace), typical
grooming verbs (put on, take off), and actions that are normally done with oneself as a recipient
or beneficiary (receive, choose, gain). Among these more agent-like middle types, the primary
figure is an affected agent, providing an alternate energy endpoint for event conception. Rather
than terminating in a distinct endpoint, the action accrues back to the primary figure, as
source/endpoint of a bidirectional transmission of force. The middle marks events that deviate
from the typical agent-patient interaction, thus highlighting the cyclic nature of the event and
profiling the direction of force that terminates with the primary figure.
4.6 Synthesis
The semantic analysis in §4.2-4.5 has implications for the description of the Greek middle.
Syntactic accounts of voice focus on describing voice as distinct from syntactic transitivity, but
from a semantic perspective, voice and transitivity, are interrelated phenomena and cannot be
considered in isolation without curtailing essential functions of each (Kemmer 1993, 247;
Shibatani 2006, 220). Discussions of voice fully rely on notions that only make sense given the
overall phenomenon of semantic transitivity in grammar, i.e. participant relations, agency,
causation, affectedness, changes of state, and salience of participants (Hopper and Thompson
1980; Mathewson and Emig 2016, 144-151; Emde Boas et al. 2019, 447-60).
By tying voice alternations to various meaning-oriented distinctions in transitivity, we
bind voice to patterns of action and interaction among participants. Viewed in this way,
transitivity represents a continuum, or gradient scale, for various semantic event types. Voice

140

relations become substantively grounded in this semantic continuum, illustrating how voice is
ultimately concerned with distinctions in event conception, or how actions unfold.
In a typical agent-patient relation, the event begins with an initiating force that is then
transferred via asymmetrical energy toward a secondary figure that goes through a process of
change. This kind of semantic event structure provides a cognitive frame for making sense of a
variety of voice alternations. Distinctions in voice represent departures from the basic transitive,
allowing for various oppositions in event structure that pertain to facets of event development
and furnish alternate ways of construing how a process may unfold. Human interactions and their
consequences provide good reason for making such distinctions in event structure.
Grammatical voice provides form-meaning relationships in language, allowing speakers
to communicate conceptual distinctions in event development, particularly in regard to their
origin, progress, and termination. The variety of middle expressions in Greek – traditionally
viewed as disparate and ill-defined – are readily organized relative to one another when framed
in terms of the developmental stages of actions. Figure 11 demonstrates a schematic depiction of
middle event types in Greek with respect to their contrast in origin, progress, and termination
from event development in the prototypical transitive.

Figure 11

175

Developmental stages of actions and Greek middle voice alternations175

Adapted from Shibatani 2006, 221.

141

Oppositions in voice embody alternative ways of viewing various facets of event
development. Passive and anticausative events illustrate a contrast in even origin, shifting the
starting point for the action, and the focus of attention, from an external, unaffected participant to
a participant internal to the process of change. Both event types involve a low-to-vanishing
saliency of an external source of energy (Kemmer 1993, 205).
For events in the cognitive domain, a contrast exists in event progress, shifting from
asymmetrical force in the active to a bidirectional transmission of force in the middle. The
experiencer is the energy source in whose mind the event originates as well as the energy
endpoint as the participant who is cognitively changed. In contrast to the active, the profiled
change occurs with the primary figure rather than the secondary participant.
Events in the cyclic domain contrast in event termination. Middle marking signals that
the source is also the endpoint of a symmetrical relation. Rather than terminating in a final
distinct participant, the event terminates where it began with an affected primary figure. On the
continuum, moving from left to right, the various middle event types are organized from those
with a more patient-like primary figure to those with a more agent-like primary figure.176

176

For a detailed semantic map of the various Hellenistic Greek middle event types, see Aubrey (2015).

142

5. Conclusion
A number of themes in our discussion of voice are worth drawing out. The first is the semantic
nature of voice alternations. One of the challenges of the middle is to understand how a variety
of different expressions are related. Historically, analysis has focused on syntactic relationships,
especially as they pertain to the role of the subject in relation to the verb. Alternations in voice
are traditionally framed as encodings of choices in clausal subject, i.e. whether or not the subject
is affected by the action. But this analytical frame is too narrow a view of event conception. It
overlooks semantic shifts in how actions unfold and allows for just one kind of contrast from the
basic active, i.e. altering the role of the subject. The present analysis frames voice in terms of
semantic transitivity, involving both semantic shifts in the type of action as well as shifts in
attentional focus regarding various facets in an event frame. Because voice resides in choices in
energy flow and attentional focus, it encompasses a range of phenomena, wherein a marked
choice in subject becomes just one type of contrast among others (Langacker 2006, 129-30). The
first has to do with changes in semantic event type with respect to how energy is transferred in an
event, particularly energy source, progress, and endpoint. The second relies on a visual metaphor
to address how cognitive attention can shift to different facets of event construal. A natural
consequence of shifting attention is a change in the relative salience of event participants.
Situating voice within event conception, especially in relation to event development and
semantic transitivity, (1) captures a variety of middle expressions in a motivated way. This
allows for voice categories like the middle to form a family of related structures, all bearing
some resemblance to one another, while also displaying different contrasts in event construal.
And (2), it points to differences in the organization of voice systems across languages. Middle
systems are largely organized around semantic distinctions in event types, whereas derived

143

passive systems rely more heavily on the role of the subject, shifting attention to the patient for
pragmatic purposes (Shibatani 2006). A middle system, like the one in Greek that subsumes both
the passive and reflexive functions in the middle-marking domain, embraces aspects of both,
with semantic and pragmatic motivations giving rise to voice alternations.
This observation about similarity and difference among voice systems brings us to the
second theme, pertaining to voice typology. The behavior of the Greek middle, and especially
the presence of middle-only verbs in the Greek voice system, may seem erratic or illogical if
considered in isolation. But when brought into the context of voice typology (chapter 2), a
number of semantic regularities are revealed that illustrate the semantic basis of middle voice
systems. Two typological patterns are central to this claim.
The first is the shared formal and semantic connections between anticausatives and
passives (§2.2.1). In Greek and other middle-marking languages, voice alternations form a
grammatical continuum of various semantic event types, subsuming the passive function and
reflexive semantics within the scope of a single form. The anticausative plays a central role as an
intervening semantic type between two extremes. It is semantically connected to the passive
through more patient-like events (e.g. σήπομαι ‘rot’) in which a patient undergoes a change of
state. It also connects to more agent-like middles through bodily actions in which an agentive
figure changes location/posture through internal energy (e.g. στρέφομαι ‘turn around’, compare
body motions to grooming middles like ἀμφιέννυμαι ‘get dressed’).
The second typological pattern is the remarkable consistency in the distribution of middle
forms across a wide variety of middle-marking languages. Middle systems are characterized by
(1) an inventory of similar semantic event types that receive middle expression, and (2) the
pervasive inclusion of non-alternating, middle-only verbs. From a primarily syntactic view of

144

voice, middle-only verbs may seem incongruous with the rest of the voice system, but there is
nothing mismatched about them once a typologically informed perspective is adopted. Greek
middle-only verbs are formally and semantically consistent with Greek as a middle voice system.
Typological patterns like these shed light on the behavior of the Greek middle by painting
a wider picture of how middle systems behave across the world’s languages. The third theme in
our discussion of voice helps to do much the same. Exploring diachronic processes (chapter 3),
contextualizes and provides motivation for particular sticking points that arise in traditional
treatments of Greek voice. Two patterns that commonly arise among middle systems occur in
Greek. Among labile alternations, the same form (active morphology), expresses both alternants.
A change in meaning occurs without a change in form: βλαστάνω ‘cause to grow’ (causative) vs.
βλαστάνω ‘bud, sprout’ (anticausative). Semantic doublets represent the opposite pattern.
Synonymous pairs (πειράω ~ πειράομαι ‘try, experience’) illustrate a change in form without a
change in meaning. Identifying lexical idiosyncrasies recognizes how such patterns arise and
where they fit within diachronic and typological patterns among middle voice systems.
In addition, processes like grammaticalization help make sense of the variety of uses that
occur with -(θ)η- morphology in Hellenistic Greek. Traditionally, -(θ)η- is treated as uniquely
passive in function, but this belies its synchronic usage and diachronic development. As in §3.2,
the integration of -(θ)η- into the Greek voice system is the result of a grammaticalization process;
a lexical-derivational form develops into a larger meaning-class via lexical expansion. Over time,
it loses specificity and expands its marking domain, culminating in a more inflectional affix,
expressing voice (middle-passive) and aspect (perfective) in the verbal system.
The development of -(θ)η- morphology from an originally change-of-state/anticausative
source, illustrates the central role of the anticausative type, not just for the rise of the -(θ)η- form,

145

but also its importance in the semantic network of middle event types. This is especially true for
ancient IE languages, e.g. Greek, Latin, and Sanskrit. Yet the anticausative use of the middle
tends to receive less attention in voice typology in comparison to the prominence given to the
more reflexive-like uses of the middle (Kemmer 1993). The anticausative is a persistent and
productive pattern in Greek voice. It warrants attention, both in regard to the diachronic origin
and rise of middle markers among languages and in regard to the semantic profile of
anticausatives in connection with the larger semantic middle domain.
Finally, chapter 4 provides a descriptive account of various semantic event types that
receive middle marking in Hellenistic Greek. These types form a semantic continuum that adopts
the scale of transitivity as a conceptual underpinning for alternations in voice (Kemmer 1993,
247). Drawing on semantic transitivity weaves voice parameters together as different facets of
the same fabric. Voice and transitivity are based in the human cognition of events and are rooted
in some of our most basic experiences of transferring action from one participant to another.
Such experiences provide the basic patterns for understanding different kinds of action. Within
event conception, these patterns pertain to developmental phases of action – how events are
brought about, how energy is transferred, and how participants are affected by the action.
Though the Greek middle may be multifunctional, it is the meaningful experiential
connections among these functions that motivate their shared formal expression, as well as their
cross-linguistic cohesion with other voice systems. It is fitting in this regard to conceive of voice
categories as a family of related structures. A set of forms in a language are used for a cluster of
related functions that then form a coherent slice of a larger network of semantic types that
languages tend to group together, based on shared conceptual distinctions in the nature of event
development and how actions unfold.

146

References
Aikhenvald, Alexandra. 2007. Typological dimensions in word formation. In Timothy Shopen
(ed.), Language typology and syntactic description, Vol. 3: Grammatical categories and
the lexicon, 1-65. Cambridge: Cambridge University Press.
Alexiadou, Artemis. 2010. On the morphosyntax of (anti-)causative verbs. In Malka Rappaport
Hovav, Edit Doron & Ivy Sichel (eds.), Syntax, lexical semantics and event structure,
177-203. Oxford: Oxford University Press.
Allan, Rutger J. 2003. The middle voice in Ancient Greek: a study in polysemy. Amsterdam: J.C.
Gieben.
Amberber, Mengistu. 2000. Valency-changing and valency-encoding devices in Amharic. In
R.M.W. Dixon & Alexandra Y. Aikhenvald (eds.), Changing valency: Cases studies in
transitivity, 312-32. Cambridge: Cambridge University Press.
Arnott, D.W. 1970. The nominal and verbal systems of Fula. Oxford: Oxford University Press.
Aubrey, Rachel E. 2015. Motivated categories, middle voice, and passive morphology. In Steven
E. Runge & Christopher J. Fresch (eds.), The Greek verb revisited: A fresh approach for
biblical exegesis, 563-635. Bellingham, WA: Lexham Press.
Bache, Carl. 1997. The study of aspect, tense, and action: towards a theory of the semantics of
grammatical categories (Rev. ed.). New York: Peter Lang.
Baerman, Matthew. 2007. Morphological typology of deponency. In Matthew Baerman, Greville
G. Corbett, Dunstan Brown, & Andrew Hippisley (eds.), Deponency and morphological
mismatches, 1-20 (Proceedings of the British Academy 145). Oxford: Oxford University.
Bauer, Laura. 2003. Morphological productivity (Cambridge Studies in Linguistics). Cambridge:
Cambridge University Press.
Beavers, John. 2011. On affectedness. Natural Language & Linguistic Theory 29(2). 335-370.
Beck, David. 2000. Unitariness of participant and event in the Bella Coola (Nuxalk) middle
voice. International Journal of American Linguistics 66(2). 218-256.
Beekes, Robert. 2010. Etymological dictionary of Greek, Vols. 1-2 (Leiden Indo-European
Etymological Dictionary Series 10/1). Leiden & Boston: Brill.
Beekes, Robert. 2011. Comparative Indo-European linguistics: An introduction (2nd ed)
(Revised by Michiel de Vaan). Amsterdam: John Benjamins.
Brinton, Laurel J. & Elizabeth C. Traugott. 2005. Lexicalization and language change (Research
Surveys in Linguistics). Cambridge: Cambridge University Press.

147

Bybee, Joan. 1985. Morphology: A study of the relation between meaning and form (Typological
Studies in Language 9). Amsterdam: John Benjamins.
Bybee, Joan. 2007. Frequency of use and the organization of language. Oxford: Oxford UP.
Bybee, Joan. 2010. Language, usage, and cognition. New York: Cambridge UP.
Campbell, Constantine R. 2015. Advances in the study of Greek: new insights for reading the
New Testament. Grand Rapids: Zondervan.
Comrie, Bernard. 1976. Aspect. Cambridge: Cambridge University Press.
Corbett, Greville G. 2007. Deponency, syncretism, and what lies between. In Matthew Baerman,
Greville G. Corbett, Dunstan Brown, & Andrew Hippisley (eds.), Deponency and
morphological mismatches, 21-44 (Proceedings of the British Academy 145). Oxford:
Oxford University Press.
Croft, William. 1993. Case marking and the semantics of mental verbs. In James Pustejovsky
(ed.), Semantics and the lexicon, 55-72. Dordrecht: Kluwer Academic.
Croft, William. 2012. Verbs: Aspect and causal structure. Oxford, UK: Oxford University Press.
Croft, William, Hava Bat-Zeev Shyldkrot, & Suzanne Kemmer. 1987. Diachronic semantic
processes in the middle voice. In Anna Giacolone Ramat, Onofrio Carruba, & Guiliano
Bernini (eds.), Papers from the 7th international conference on historical linguistics,
179-92. Amsterdam: John Benjamins.
Croft, William and D. Alan Cruse. 2004. Cognitive linguistics. Cambridge: Cambridge UP.
Crystal, David. 2003. An encyclopedia of linguistics and phonetics (5th ed.). Oxford: Blackwell.
Dana, H.E. & Julius R. Mantey. 1927. A manual of the Greek New Testament. New York:
Macmillan.
Decker, Rodney J. 2014. Reading koine Greek: an introduction and integrated workbook. Grand
Rapids: Baker Academic.
Dench, Alan. 2013. Possession in Martuthunira. In Alexandra Y. Aikhenvald & R. M. W. Dixon
(eds.), Possession and ownership: A cross-linguistic typology. Oxford: Oxford UP.
Dixon, R.M.W. & Alexandra Y. Aikhenvald (eds.). 2000. Changing valency: Cases studies in
transitivity. Cambridge: Cambridge University Press.
Ellis, Nicholas J. 2016. Aspect-prominence, morpho-syntax, and a cognitive-linguistic
framework for the Greek verb. In Steven E. Runge & Christopher J. Fresch (eds.), The
Greek verb revisited: A fresh approach for biblical exegesis, 122-60. Bellingham, WA:
Lexham Press.

148

Emde Boas, Evert van, Albert Rijksbaron, Luuk Huitink, & Mathieu de Bakker 2019. Cambridge
grammar of Classical Greek. Cambridge: Cambridge University Press.
Faltz, Leonard M. 1977. Reflexivization: a study in universal syntax. Doctoral dissertation,
University of California, Berkeley (Reprinted 2017 by Routledge, London).
Farrell, Patrick. 2005. Grammatical relations. Oxford: Oxford University Press.
Foley, William A. & Robert D. Van Valin, Jr. 1984. Functional syntax and universal grammar.
Cambridge: Cambridge University Press.
Fox, Barbara & Paul J. Hopper (eds.). 1994. Voice form and function (Typological Studies in
Language 27). Amsterdam & Philadelphia: John Benjamins.
Frazer, Jennifer. 2015. Dying trees can send food to neighbors of different species.
https://blogs.scientificamerican.com/artful-amoeba/dying-trees-can-send-food-toneighbors-of-different-species (9 June 2018).
García Ramón, José Luis. 2014. From Aktionsart to aspect and voice: On the morphosyntax of
the Greek aorists in -η- and -(θ)η-. In Annamaria Bartolotta (ed.), The Greek verb:
Morphology, syntax, and semantics: Proceedings of the 8th international meeting of
Greek linguistics (Agrigento, October 1-2, 2009) [Bibliothèque des Cahiers de
Linguistique de Louvain (BCLL) 128], 149-82. Leuven: Peeters.
Geniušienė, Emma. 1987. The typology of reflexives (Empirical Approaches to Language
Typology 2). Berlin: Mouton de Gruyter.
Gerdts, Donna B. & Thomas E. Hukari. 2006. The Halkomelem middle: A complex network of
constructions. Anthropological Linguistics 48(1). 44-81.
Gildersleeve, Basil Lanneau. 1900. Syntax of classical Greek from Homer to Demosthenes. New
York: American Book Company.
Givón, Talmy. 1984. Syntax: a functional-typological introduction. Vol. 1. Amsterdam: John
Benjamins.
Givón, Talmy. 2002. Bio-linguistics: The Santa Barbara lectures. Amsterdam: John Benjamins.
Goodwin, W. Watson. 1895. Greek reader: consisting of selections from Xenophon, Plato,
Herodotus, and Thucydides, with notes adapted to the revised and enlarged edition of
Goodwinʹs Greek grammar, and copperplate maps (Rev. ed.). Boston: Ginn & Company.
Harris, Roy. 1998. Making sense of communicative competence. In Roy Harris & George Wolf
(eds.), Integrational linguistics: a first reader, 27-45 (Language & Communication
Library 18). Oxford: Pergamon.

149

Harris, Roy & George Wolf (eds.). 1998. Introduction to Integrational linguistics: a first reader
(Language & Communication Library 18). Oxford: Pergamon.
Haspelmath, Martin. 1987. Transitivity alternations of the anticausative type. Institut für
Sprachwissenschaft Arbeitspapier NS 5. Cologne: Universität zu Köln.
Haspelmath, Martin. 1990. The grammaticization of passive morphology. Studies in Language
14(1). 25-72.
Haspelmath, Martin. 1993. More on the typology of inchoative/causative verb alternations. In
Bernard Comrie & Maria Polinksy (eds.), Causatives and transitivity, 87-120.
Amsterdam: John Benjamins.
Haspelmath, Martin. 2003. The geometry of grammatical meaning: semantic maps and crosslinguistic comparison. In Michael Tomasello (ed.), The new psychology of language:
cognitive and functional approaches to language structure, vol. 2, 211-42. Mahwah, NJ:
Lawrence Erlbaum.
Haspelmath, Martin. 2016. Universals of the causative and anticausative verb formation and the
spontaneity scale. Lingua Posnaniensis 58(2). 33-63.
Haspelmath, Martin & Thomas Müller-Bardey. 2004. Valency change. In Geert Booij, Christian
Lehmann, Joachim Mugdan, & Stavros Skopeteas (eds.), Morphology: An international
handbook on inflection and word-formation, Vol. 2, 1130-44. Berlin: Walter de Gruyter.
Hinge, George. 2007. The PIE essive and fientive in Greek (handout), Greek and Latin from an
Indo-European Perspective II, University of Oslo, 5-7 July 2007.
http://www.glossa.dk/lingu/essiv.pdf (11 November 2019).
Hock, Hans Henrich. 2003. Analogical change. In Brian D. Joseph & Richard D. Janda (eds.),
The handbook of historical linguistics, 441–460. Oxford: Blackwell.
Hopper, Paul J. 1998. Emergent Grammar. In Michael Tomasello (ed.), The new psychology of
language: cognitive and functional approaches to language structure, vol. 1, 155-75.
Mahwah, NJ: Lawrence Erlbaum.
Hopper, Paul J. & Sandra A. Thompson. 1980. Transitivity in grammar and discourse. Language
56(2): 251–299.
Jassanoff, Jay. 2002/2003. Stative *-ē-revisited. Die Sprache 43. 127-70.
Junčytė, Giedrė. 2018. Lithuanian anticausative verb triads: Form, semantics and functions. In:
Andra Kalnača and Ilze Lokmane (eds.), Language: Meaning and Form 9, Grammar and
pragmatics, 80-95. Latvijas Universitāte.

150

Keenan, Edward L. & Matthew S. Dryer. 2007. Passive in the world’s languages. In Timothy
Shopen (ed.), Language typology and syntactic description, Vol. 1: Clause structure (2nd
ed.), 325-61. Cambridge: Cambridge University Press.
Kemmer, Suzanne. 1993. The middle voice (Typological Studies in Language 23). Amsterdam &
Philadelphia: John Benjamins.
Kemmer, Suzanne. 2003. Human cognition and the elaboration of Events: some universal
conceptual categories. In Michael Tomasello (ed.), The new psychology of language:
cognitive and functional approaches to language structure, vol. 2, 89-118. Mahwah, NJ:
Lawrence Erlbaum.
Kemmer, Suzanne & Arie Verhagen. 1994. The grammar of causatives and the conceptual
structure of events. Cognitive Linguistics 5(2). 115-56.
Kiefer, Ferenc. 2000. Regularity. In Geert Booij, Christian Lehmann, Joachim Mugdan, &
Stavros Skopeteas (eds.), Morphology: An international handbook on inflection and
word-formation, Vol. 1, 296-302. Berlin: Walter de Gruyter.
Klaiman, M.H. 1991. Grammatical voice. New York: Cambridge University Press.
Krasukhin, Konstantin G. 2006. Typology and comparative linguistics: Jakobson revisited. In
Terttu Nevalainen, Juhani Klemola, & Mikko Laitinen (eds.), Types of variation:
Diachronic, dialectal and typological interfaces (Studies in Language Companion Series
76), 81-97. Amsterdam & Philadelphia: John Benjamins.
Kretzschmar, William A, Jr. 2015. Language and complex systems. Cambridge: Cambridge UP.
Kulikov, Leonid. 1998. Passive, anticausative and classification of verbs: The case of Vedic. In
Leonid Kulikov & Heinz Vater (eds.), Typology of verbal categories: Papers presented
to Vladimir Nedjalkov on the occasion of his 70th birthday, 139-54. Tübingen: Niemeyer.
Kulikov, Leonid. 2006. Passive and middle in Indo-European: Reconstructing the early Vedic
passive paradigm. In Werner Abraham and Larisa Leisiö (eds.), Passivization and
typology: Form and function, 62-81. Amsterdam: John Benjamins.
Kulikov, Leonid. 2010. Bridging typology and diachrony: A preliminary questionnaire for a
diachronic typological study of voice and valency-changing categories. In Valentin
Vydrin et al. (eds.), Problemy grammatiki i tipologii: sbornik statej pamjati Vladimira
Petroviča Nedjalkova (1928–2009) [Issues in Grammar and Typology: A Memorial Volume
for Vladimir Nedjalkov], 139-63. Moscow: Znak.
Kulikov, Leonid. 2011. Voice typology. In Jae Jung Song (ed.), The oxford handbook of
linguistic typology, 368-98. Oxford: Oxford University Press.
Kulikov, Leonid. 2013. Middle and reflexive. In Silvia Luraghi & Claudia Parodi (eds.), The
bloomsbury companion to syntax, 261-280. New York: Bloomsbury.

151

Kozinsky, Isaac and Vladimir Nedjalkov, & Maria Polinskaja. 1988. Antipassive in Chukchee.
In Masayoshi Shibatani (ed.), Passive and voice, 651-706. Amsterdam: John Benjamins.
Ladewig, Stratton L. 2010. Defining deponency: An investigation into Greek deponency of the
middle and passive voices in the koine period. Doctoral dissertation, Dallas Theological
Seminary.
Lakoff, George. 1987. Women, fire, and dangerous things: What categories reveal about the
mind. Chicago: University of Chicago Press.
Langacker, Ronald W. 1987. Foundations of cognitive grammar: Vol. 1 Theoretical
prerequisites. Stanford, CA: Stanford University Press.
Langacker, Ronald W. 1991. Foundations of cognitive grammar: Vol. 2 descriptive application.
Stanford, CA: Stanford University Press.
Langacker, Ronald W. 1998. Conceptualization, symbolization, and grammar. In Michael
Tomasello (ed.), The new psychology of language: cognitive and functional approaches
to language structure vol. 1, 1-39. Mahwah, NJ: Lawrence Erlbaum.
Langacker, Ronald W. 2006. Dimensions of defocusing. In Tasaku Tsunoda & Taro Kageyama
(eds.), Voice and grammatical relations In honor of Masayoshi Shibatani, 115-37
(Typological Studies in Language 65). Amsterdam & Philadelphia: John Benjamins.
Langacker, Ronald W. 2010. How not to disagree: The emergence of structure and usage. In
Kasper Boye & Elisabeth Engberg-Pedersen (eds.), Language usage and language
structure, 107-43 (Trends in Linguistics, Studies and Monographs 213). Berlin & New
York: Mouton de Gruyter.
LaPolla, Randy J. 2003. Why languages differ: variation in the conventionalization of constraints
on inference. In David Bradley, Randy J. LaPolla, Boyd Michailovsky & Graham
Thurgood (eds.), Language variation: papers on variation and change in the sinosphere
and in the indosphere in honour of James A. Matisoff, 113-44 (Pacific Linguistics 555).
Canberra: The Australian National University.
LaPolla, Randy J., František Kratochvíl & Alexander R. Coupe. 2011. On transitivity. Studies in
Language 35(3). 469-491.
Lavidas, Nikolaos and Dimitra Papangeli. 2007. Deponency in the diachrony of Greek. In
Matthew Baerman, Greville G. Corbett, Dunstan Brown, & Andrew Hippisley (eds.),
Deponency and morphological mismatches, 97-126 (Proceedings of the British Academy
145). Oxford: Oxford University Press.
Lazard, Gilbert. 2003. Transitivity revisited as an example of a more strict approach in
typological research. Folia Linguistica XXXVI(3-4). 141–190.

152

Levin, Beth. 1993. English verb classes and alternations: A preliminary investigation. Chicago:
University of Chicago Press.
Levin, Beth & Malka Rappaport Hovav. 1995. Unaccusativity at the syntax-lexical semantics
interface. Cambridge, Mass.: MIT Press.
Lyons, John. 1968. Introduction to theoretical linguistics. New York: Cambridge University.
Maldonado, Ricardo. 2007. Grammatical voice in cognitive grammar. In Dirk Geeraerts &
Hubert Cuyckens (eds.), The oxford handbook of cognitive linguistics, 829-68. Oxford:
Oxford University Press.
Maldonado, Ricardo. 2008. Spanish middle syntax. A usage-based proposal for grammar
teaching. In Sabine De Knop & Teun De Rycker (eds.), Cognitive approaches to
pedagogical grammar, 155-96. Berlin: Moutin de Gruyter.
Maldonado, Ricardo. 2009. Middle as a basic voice system. In Lilián Guerrero, Sergio Ibáñez
Cerda, & Valeria A. Belloro (eds.), Studies in role and reference grammar, 69-109.
Mexico: Instituto de Investigaciones Filológicas, UNAM.
Manney, Linda Joyce. 2000. Middle voice in Modern Greek (Studies in Language Companion
Series 48). Amsterdam: John Benjamins.
Manzini, M. Rita, Anna Roussou, & Leonardo M. Savoia. 2016. Middle-passive voice in
Albanian and Greek. Journal of linguistics 52(1). 111-150.
Matthews, Peter H. 2007. How safe are our analyses?. In Matthew Baerman, Greville G. Corbett,
Dunstan Brown, & Andrew Hippisley (eds.), Deponency and morphological mismatches,
297-315 (Proceedings of the British Academy 145). Oxford: Oxford University Press.
Mathewson, David L. & Elodie Ballantine Emig. 2016. Intermediate Greek grammar: syntax for
students of the new testament. Grand Rapids: Baker Academic.
Martin, Jack B. 2000. Creek voice: beyond valency. In R.M.W. Dixon & Alexandra Y.
Aikhenvald (eds.), Changing valency: Cases studies in transitivity, 375-402. Cambridge:
Cambridge University Press.
Mithun, Marianne. 2006. Voice without subjects, objects, or obliques: Manipulating argument
structure in Agent/Patient systems (Mohawk). In Tasaku Tsunoda & Taro Kageyama
(eds.), Voice and grammatical relations: in honor of Masayoshi Shibatani, 195-216
(Typological Studies in Language 65). Amsterdam & Philadelphia: John Benjamins.
Mounce, William D. 2009. Basics of Biblical Greek grammar (3rd ed.). Grand Rapids:
Zondervan.
Næss, Åshild. 2007. Prototypical transitivity (Typological Studies in Language 72). Amsterdam
& Philadelphia: John Benjamins.

153

Nedjalkov, Vladimir P. 2007a. Encoding of the reciprocal meaning. In Vladimir P. Nedjalkov,
Emma Š. Geniušienė, & Zlatka Guentchéva (eds.), Reciprocal constructions, 147-207
(Typological Studies in Language 71). Amsterdam & Philadelphia: John Benjamins.
Nedjalkov, Vladimir P. 2007b. Polysemy of reciprocal markers. In Vladimir P. Nedjalkov,
Emma Š. Geniušienė, & Zlatka Guentchéva (eds.), Reciprocal constructions, 231-334
(Typological Studies in Language 71). Amsterdam & Philadelphia: John Benjamins.
Nedjalkov, Vladimir P. 2007c. Overview of the research. Definitions of terms, framework, and
related issues. In Vladimir P. Nedjalkov, Emma Š. Geniušienė, & Zlatka Guentchéva
(eds.), Reciprocal constructions, 3-114 (Typological Studies in Language 71).
Amsterdam & Philadelphia: John Benjamins.
Nedjalkov, Vladimir P. & Emma Geniušienė. 2007. Questionnaire on reciprocals. In Vladimir P.
Nedjalkov, Emma Š. Geniušienė, & Zlatka Guentchéva (eds.), Reciprocal constructions,
379-434 (Typological Studies in Language 71). Amsterdam: John Benjamins.
Nichols, Johanna. 2003. Diversity and stability in language. In Brian D. Joseph & Richard D.
Janda (eds.), The handbook of historical linguistics, 283-310. Berlin: Blackwell.
Onishi, Masayuki. 2000. Transitivity and valency-changing derivations in Motuna. In R.M.W.
Dixon & Alexandra Y. Aikhenvald (eds.), Changing valency: Cases studies in
transitivity, 115-44. Cambridge: Cambridge University Press.
Palancar, Enrique L. 2004. Middle voice in Otomi. International Journal of American
Linguistics 70(1). 52-85.
Pavey, Emma. 2010. The structure of language. Cambridge: Cambridge University Press.
Payne, Thomas E. 2006. Exploring language structure: A student’s guide. Cambridge:
Cambridge University Press.
Porter, Stanley E. 1994. Idioms of the Greek new testament (2nd ed.). Sheffield: Sheffield
Academic.
Porter, Stanley E., Jeffrey T. Reed, & Matthew Brook O’Donnell. 2010. Fundamentals of New
Testament Greek. Grand Rapids, MI: William B. Eerdmans Publishing Company.
Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, Jan Svartvik. 1985. A comprehensive
grammar of the English language. New York: Longman Group Limited.
Richards, Jack C. & Theodore S. Rodgers. 2014. Approaches and methods in language teaching
(3rd ed.). Cambridge: Cambridge University Press.
Rijksbaron, Albert. 2006. The syntax and semantics of the verb in classical Greek: An
introduction (3rd ed.). Chicago: University of Chicago Press.

154

Robertson, Archibald T. 2006 [1919]. Grammar of the Greek New Testament in the light of
historical research. Bellingham, WA: Faithlife.
Saeed, John I. 1995. The semantics of middle voice in Somali. African Languages and Cultures
8(1). 61-85.
Sansò, Andrea. 2006. ‘Agent defocusing’ revisited: Passive and impersonal constructions in
some European languages. In Werner Abraham & Larisa Leisiö (eds.), Passivization and
typology: Form and Function (Typological Studies in Language 68), 232-273.
Amsterdam: John Benjamins.
Sauer, Hans. 2004. Lexicalization and demotivation. In Geert Booij, Christian Lehmann,
Joachim Mugdan, & Stavros Skopeteas (eds.), Morphology: An international handbook
on inflection and word-formation, Vol. 2, 1625-1635. Berlin: Walter de Gruyter.
Sausa, Eleonora. 2016. Basic valency orientation in Homeric Greek. Folia Linguistica Historica
37. 205-238.
Shibatani, Masayoshi. 1985. Passives and related constructions: A prototype analysis. Language
61, 821–848.
Shibatani, Masayoshi (ed.). 1988. Passive and voice (Typological Studies in Language 16).
Amsterdam: John Benjamins.
Shibatani, Masayoshi. 2004. Voice. In G. Booij et al. (eds.), Morphology: A handbook on
inflection and word-formation, Vol. 2, 1145-65. Berlin: Walter de Gruyter.
Shibatani, Masayoshi. 2006. On the conceptual framework for voice phenomena. Linguistics 44.
217–269.
Shibatani, Masayoshi and Ketut Artawa. 2007. The middle voice in Balinese. In Iwasaki Shoichi,
Andrew Simpson, Karen Adams, & Paul Sidwell (eds.), SEALS XIII: Papers from the
13th Annual Meeting of the Southeast Asian Linguistics Society, 239-261. Canberra,
Australia: Pacific Linguistics, Research School of Pacific and Asian Studies.
Sihler, Andrew L. 1995. New comparative grammar of Greek and Latin. New York: Oxford.
Simard, Suzanne W. 2018. Mycorrhizal networks facilitate tree communication, learning, and
memory. In Frantisek Baluska, Monica Gagliano, & Guenther Witzany (eds.), Memory
and learning in plants, 191-213. New York: Springer.
Smyth, Herbert Weir. 1956. Greek grammar (Rev. ed.). Cambridge: Harvard University Press.
Steinbach, Markus. 2002. Middle voice: A comparative study in the syntax-semantic interface of
German (Linguistik Aktuell/Linguistics Today 50). Amsterdam: John Benjamins.

155

Sweetser, Eve. 1990. From etymology to pragmatics: Metaphorical and cultural aspects of
semantic structure. Cambridge: Cambridge University Press.
Szemerényi, Oswald. 1996. Introduction to Indo-European linguistics. Oxford: Oxford UP.
Talmy, Leonard. 1985. Lexicalization patterns: Semantic structure in lexical forms. In Timothy
Shopen (ed.), Language typology and syntactic description, Vol. 3: Grammatical
categories and the lexicon. Cambridge, Cambridge University Press.
Tsunoda, Tasaku & Taro Kageyama (eds.). 2006. Voice and grammatical relations: In honor of
Masayoshi Shibatani (Typological Studies in Language 65). Amsterdam: John
Benjamins.
Van Valin, Robert D., Jr. & Randy J. LaPolla. 1997. Syntax: structure, meaning, and function.
Cambridge: Cambridge University Press.
Verhagen, Arie. 2002. From parts to wholes and back again. Cognitive Linguistics 13(4). 403-39.
Verhagen, Arie. 2007. Construal and perspectivization. In Dirk Geeraerts & Hubert Cuyckens
(eds.), The oxford handbook of cognitive linguistics, 48-81. Oxford: Oxford University.
Wackernagel, Jacob. 2009. Lectures on syntax: with special reference to Greek, Latin, and
Germanic (David R. Langslow, ed.). Oxford: Oxford University Press.
Wallace, Daniel B. 1996. Greek grammar beyond the basics: an exegetical syntax of the new
testament. Grand Rapids: Zondervan.
Ward, Gregory, Betty Birner, & Rodney G. Huddleston. 2002. Information packaging. In
Rodney D. Huddleston & Geoffrey K. Pullum (eds.), The Cambridge grammar of the
English language, 1363-1448. New York: Cambridge University Press.
Wiemer, Björn and Vladimir P. Nedjalkov. 2007. Reciprocal and reflexive constructions in
German. In Vladimir P. Nedjalkov, Emma Š. Geniušienė, & Zlatka Guentchéva (eds.),
Reciprocal constructions, 455-512 (Typological Studies in Language 71). Amsterdam &
Philadelphia: John Benjamins.
Willi, Andreas. 2018. Origins of the Greek verbs. Cambridge: Cambridge University Press.
Wolde, Ellen van. 2019. The Niphal as middle voice and its consequence for meaning. Journal
for the Study of the Old Testament 43(3). 453-478.
Young, Richard A. 1994. Intermediate New Testament Greek: A linguistic and exegetical
approach. Nashville, Tenn: Broadman & Holman.