DOI: https://d0i.0rg/10.4312/keria.22.2.7-23
Geoffrey Horrocks
What's in the Middle?
Two Voices or Three in Ancient Greek?
1. INTRODUCTION
When students start to learn Ancient Greek, they quickly learn that the language has three grammatical voices, active, middle and passive, which in different ways articulate the relationship between grammatical functions like subject, direct object and indirect object, and semantic roles like agent, patient and experiencer. The three voices are functionally characterised by Allan (2014) in the short abstract that begins his article on Voice in the online Encyclopedia of Greek Language and Linguistics:
While the active voice is semantically unmarked, the middle voice expresses that the subject is affected. The passive voice indicates that the subject is a fully affected patient/theme or experiencer.
The immediate problem in Allan's characterisation is the absence of a sharply defined contrast between the middle and the passive. Since the middle and passive are formally distinct only in the aorist and future, and then only in part (see §2 below), they are in fact treated as a single but polysemous "mediopassive voice" indicating varying degrees of the "affectedness" of the subject. But this approach obscures a fundamental difference between the passive and the middle which will now be explored.
On the one hand, the active-passive relationship is highly regular and productive in that sentences containing active transitive verbs almost always have intransitive passive counterparts regardless of the lexical meaning of the verbs involved. This is, in other words, an essentially syntactic relationship with predictable structural and semantic effects, as summarised in (1), where the agent
8
Geoffrey Horrocks
of the active sentence has been downgraded to the status of optional adjunct in the passive counterpart, and the patient of the active sentence has become the subject of the passive one:
(1)	subject - active verb - object < > subject - passive verb (- by-phrase)
agent predicate patient	patient predicate	agent
The fanatics burned the books.	The books were burned (by the fanatics).
The same situation can therefore be described in two different ways—or equivalently, in two different grammatical voices.
By contrast, it is much more difficult to characterise the middle voice (even the name is vague, implying a function of unspecified nature between those of the active and passive). This is because its function is neither regular nor predictable. The term is typically employed in general linguistics to cover a range of detransitivisation processes that have effects similar to those of the passive, but with some crucial differences. Consider first the English examples in (2):
(2)	(a) This essay reads beautifully.	core "middle" use of a verb
(b)	Max washed/shaved/dressed (i.e., himself). implicit reflexive use of a verb
(c)	The door is closing.	anticausative use of a verb
In each case a normally active transitive verb is used intransitively, but now, as in the passive, the subject denotes the theme or patient of the action, whether exclusively, as in (2a) and (2c), or in combination with the agent, as in (2b). An external agent may be implied in both (2a) and (2c), but this cannot be identified with a fry-phrase: e.g. *this essay reads beautifully by the professor is unacceptable. Notice that (2a), the type specifically identified as "middle", normally requires some form of adverbial modification to be grammatical: e.g. *this essay reads is unacceptable. This is not true of (2b) and (2c), where the verb can stand alone. It is also important to note that (2c) involves an alternation between a specifically causative transitive verb and an intransitive counterpart with a theme/patient subject (an "anticausative" or "unaccu-sative"): e.g. verbs like break, melt, boil, freeze, open, close, burn. These verbs normally involve a change of state (or sometimes location), so that the transitive verb means 'X causes Y to become Z', and the intransitive verb means 'Y becomes Z'. The three types in (2) have much in common, and are often treated together as phenomena characteristic of the "middle voice". Indeed, it can be difficult in specific cases to distinguish clearly among them, as (3) makes clear:
What's in the Middle? Two Voices or Three in Ancient Greek?
9
(3) This program - downloads quickly.
-	has downloaded (i.e. itself).
-	is downloading (i.e. automatically).
(?middle)
(?implicit reflexive) (?anticausative)
But there is one critical difference between the active-passive relation and the active-middle relation: where the former is fully productive (sentences with active transitive verbs almost always have passive counterparts regardless of their meaning), the latter is lexically highly restricted: only certain transitive verbs, or transitive verbs with certain types of meaning, allow for intransitive middle uses alongside their active transitive use, as the ungrammatical examples in (4) show:
(4) (a) *These fixtures destroy/design easily. impossible as "middles"
(b)	*Max hit/amused.	impossible as implicit reflexives
(c)	*My essay is writing/researching.	impossible as anticausatives
In other words, it makes little sense to view the active-middle relationship as a structural one comparable to the active-passive one when the existence and meaning of a middle counterpart is determined not by general syntactic properties but by specific lexical ones. If the middle voice can be characterised in a coherent way at all, it would clearly be better to try to capture its essence by means of lexical rules that affect only the relevant sub-classes of verbs.
Mutatis mutandis, the conglomeration of properties discussed for English middles typically recurs cross-linguistically, even though the resulting middle voice may be realised in different ways. Accordingly, reflexivity and a range of other non-active/non-passive functions have traditionally been grouped together as "middle" in modern grammars of Ancient Greek (most recently, van Emde Boas, Rijksbaron, Huitink, and de Bakker 2019, Chap. 35) (but see also the discussions in Allan 2003, 2014 and Kemmer 1993, and the articles in Fox and Hopper 1994, especially those by Bakker, Givon and Yang, and Kemmer). This approach contrasts strongly with the ancient grammatical tradition (Dionysius Thrax, Heliodorus, Apollonius Dyscolus, Choero-boscus), which struggled to find any obvious rationale for the middle voice and treated it largely as a dustbin for formal and functional oddities that were neither clearly active nor clearly passive (see Rijksbaron 2018 for a thorough treatment). One major purpose of this article, then, is to try to answer the question of which tradition is closer to the truth: did Ancient Greek really
2. VOICE(S) IN ANCIENT GREEK
10
Geoffrey Horrocks
have three voices, or just two, with some residual data that cannot readily be classified as either?
As we have seen, English uses active verb forms to express typical middle meanings, but other languages may use passive or reflexive forms in the same range of functions. It is very rare, however, for a middle voice to have a distinctive morphology of its own. Thus, as noted above, Ancient Greek middle and passive verb forms largely coincide, as the umbrella term "medio-passive" implies. But even where there is in theory a formal distinction, specifically in the aorist and the future, there is in practice a great deal of overlap, with no consistent correlation of form and function. For example, there are verbs with morphologically middle futures used in a passive sense (e.g., ti|iao|iai 'I shall be honoured', ^avou^ai 'I shall be shown'), and many verbs with morphologically passive aorists used in a middle sense alongside morphologically middle futures. Some common examples of the latter are given in (5):
(5) Middle verbs with the supposedly "passive" aorist -(6)r|v but a middle future: £PouAr|9r|v/pouAr|ao|iai 'wish/want, sSuvr|6r|v/Suvr|ao|iai 'be able', anr|AAd-yr|v/ anaAAd^o|iai 'depart, SKivr|0r|v/Kivr|ao|iai 'move, sAunr|0r|v/AuTCr|ao|iai 'grieve'
In the "modern" approach, the Greek medio-passive paradigm is typically seen as a polysemous marker of the "affectedness" of a subject, i.e. the agentive subject of an active verb is reinterpreted as receiving, either additionally (middle) or instead (passive), the "effect" of the verbal action as a theme or patient. A possible path for the semantic development of detransitivised medio-passive functions is given through the English examples in (6):
(6) (a)	Socrates beat his wife	agent only	active verb
(b)	Socrates dressed his son	agent only	active verb
(c)	Socrates got (himself) beaten	(indirect agent+) patient	passive verb
(d)	Socrates got (himself) dressed	direct agent+patient	middle verb
(e)	Socrates got beaten by his wife	patient only	passive verb
(f)	*Socrates got dressed by his wife	*patient + direct agent	*middle verb
(g)	Socrates was beaten by his wife	patient only	passive verb
(h)	Socrates was dressed by his wife	patient only	passive verb
(6a) and (6b) contain the active transitive verbs, beat and dress. Let us suppose for the sake of argument that the English get + passive participle construction in (6c) and (6d) corresponds functionally to Greek medio-passive morphology, and that it contributes a nuance of "reflexivity" to actions proto-typically involving agentive subjects. This may be overtly expressed by means
What's in the Middle? Two Voices or Three in Ancient Greek?
11
of a reflexive pronoun, or be implicitly understood (as indicated by the brackets around himself). There is, however, a crucial difference between (6c) and (6d) determined by lexical semantics. Beat is an activity that normally involves distinct agents and patients (i.e., people don't usually beat themselves), while dress readily allows for agents to act on themselves (i.e., people do normally dress themselves). So (6c) with the reflexive pronoun means that Socrates did something that caused someone else to beat him, while (6d) with the reflexive pronoun simply means that Socrates dressed himself: i.e. the first involves indirect agency, the second direct agency, with respect to the relevant activity. Accordingly, (6c) allows for "a beater" to be specified, cf. (6e), while (6d) does not permit the specification of "a dresser" other than Socrates, cf. (6f). But when the reflexive pronoun is dropped in these examples, the meaning of (6c) changes while that of (6d) stays the same: specifically, the idea that Socrates was somehow indirectly responsible for his own beating disappears along with the reflexive pronoun, but the idea that he dressed himself remains. We may conclude, then, that (6c), with or without the reflexive pronoun, is passive, but that (6d), with or without the reflexive pronoun, is middle. In the case of verbs with meanings like "dress" a true passive reading is only possible when the sense of direct agency is unambigously removed through the substitution of be for get: cf. (6h), where a distinct agent has been added successfully. For verbs with meanings like 'beat', however, the two auxiliaries are more or less interchangeable in passive function, as shown by (6e) and (6g), though the former but not the latter suggests that Socrates was also something of an experiencer as well as a mere (inert) patient.
There are, however, other transitive verbs, including those with corresponding "core middle" or anticausative uses, that allow for both passive and middle readings of the get-construction. In this case, we either understand that the action was performed by an external agent on the patient subject, as in (7a) and (7c), or that it occurred more or less spontaneously, as a result of some inherent property of the patient subject and/or the ambient circumstances, as in (7b) and (7d):
(7) (a) This clay gets moulded quite easily (e.g., by a skilled potter) passive
(b)	This clay gets moulded quite easily (i.e., all by itself)	middle
(c)	The wax got melted (e.g., by the flames)	passive
(d)	The wax got melted (i.e., all by itself)	middle
Unlike in (6c) and (6d), therefore, the patient subject here is not, strictly speaking, also an agent, though it still plays a residually "active" kind of role because of its inherent properties, and reflexive pronouns may be marginally allowed (cf. this clay gets ?itself moulded quite easily etc.).
12
Geoffrey Horrocks
This kind of explanatory framework can be adapted and summarised for Ancient Greek as in (8):
(8)	(a) any active transitive verb may take medio-passive morphology in passive
function and co-occur optionally with an agentive phrase (uno + genitive etc.) (b) any active transitive verb with the appropriate lexical semantics may also take medio-passive morphology in a middle function, but cannot then co-occur with an agentive phrase
Thus implicit reflexives, for example, are largely restricted to a small number of verbs denoting activities involving personal grooming and training: e.g. Xouw/Xouo|iai 'wash', YU|vdZ«/YU|ivdZo|iai 'train, etc. However, the kind of function associated with the core middles in English is typically performed by Greek verb forms that are just as likely to be passive as middle in force, as in (9):
(9)	(a) This clay moulds easily.
(b) oOto; o nr|A.6; paSi«; nXaTxexai (? = 'is moulded easily' (sc. by anyone at all))
And the relatively large class of verb forms corresponding to English anti-causatives may also be passive in sense, as in (10):
(10)	(a) The wax melted.
(b) o K^po; £T&Kr| (? = 'was melted' (sc. by unknown factors))
In other words, since both these classes can in principle co-occur with agentive or instrumental phrases, we have no way of knowing in the absence of native speakers whether there was also a distinct middle reading (= 'moulds easily/melted—all by itself') that rejected such an addition. The conclusion that these forms may well be universally passive is reinforced by the fact that there are good examples of active anticausatives, as the verbs of movement in (11):
(11)	sAauv« 'drive/proceed, op|i« '(cause to) start out', aneuS« '(cause to) hasten, unay« 'withdraw/go'
Accordingly, this overall state of affairs potentially leaves the set of mediopassive verb forms with clear middle meanings perilously small. Traditional grammars boost the numbers, however, by including transitive middles. Unlike the data typically discussed as middles in the general linguistic context, large numbers of formally middle verbs in Greek are in fact transitive rather than intransitive, and have a specifically "middle" aorist in -(a)d|ir|v or -o|ir|v that is rarely, if ever, passive/intransitive in meaning. This is clearly a novel
What's in the Middle? Two Voices or Three in Ancient Greek?
13
type of middle in that all the examples discussed so far, in both English and Greek, have been intransitive, and as such closer to passives than to active transitives. We might speculate, faute de mieux, that the more common transitive type of middle may have been formed prehistorically by analogy with the type of middle exemplified by verbs of personal grooming and training such as Xouw/Xouo|ai, "yu|vdZw/"yu|vdZo|ai, etc. Consider (12):
(12)	(a) Xouw £|iauxov : Xouo|iai 'I wash myself'
subject = agent+patient (b) noiu Ti E|iauT« : > noiou|iai Ti 'I make something for myself'
subject = agent+beneficiary
This analogy would have been based on the assumption that an active verb co-occurring with an overt reflexive pronoun could be replaced by a middle verb form with reflexive meaning, whether the reflexive in question was a direct object or an indirect object. We would therefore end up with implicit direct reflexives expressed by intransitive middles, as in (12a), and a new class of of implicit indirect reflexives expressed by transitive middles, as in (12b).
Since the set of active transitive verbs that can in principle co-occur with a dative object or adjunct (denoting a recipient, an experiencer, a beneficiary, etc.) is quite large, the set of associated transitive middles should therefore be correspondingly large, at least in theory. Much is made of this in modern grammars and lexica, where the transitive middle is typically said to denote an action that an agent performs "for himself/herself/itself", though sometimes vaguer versions of indirect reflexivity are also invoked. The only problem with this statement is that it simply is not true. Note first of all that the only example of the construction that is ever discussed in these terms in the ancient grammatical tradition (scholion on Heliodorus 1.3.246.5 [= part 1, volume 3, page 246, line 5 in Grammatici Graeci, edited by Uhlig-Schneider-Hilgard]) is precisely the one in (12b), albeit presented there in the aorist. If things were really so clear and simple, this would surely have been developed as the basis for a reasoned theory of the transitive middle. The fact that it was not speaks volumes. In reality, the supposedly straightforward indirect-reflexive sense of a transitive middle is rare, being restricted to a relatively small set of semanti-cally linked verbs, including those in (13):
(13)	noi«/noiou|iai 'make', napaCTKeudZ«/napapaCTKeudZo|ai 'prepare', TCapex«/napexo|iai 'provide'
This limitation is not difficult to explain. Since people frequently and naturally "make", "prepare" or "provide" things for themselves, the lexical
14
Geoffrey Horrocks
meaning of these verbs strongly invites a direct agent reading of the subject of their middle forms, e.g. napapaaK£ud(o|ai = 'get something prepared (for one's own benefit/use)' etc. As we saw above in (6d), this particular interpretation of the subject is a prerequisite for the possibility of a true reflexive reading of the get-paraphrase. If instead the agent is understood to be acting indirectly, as was the case in (6c), the possibility of reflexivity is eliminated and the reading is a simple causative one, cf. "Socrates got his wife beaten" (= 'caused his wife to be beaten'). But even when a subject can be understood as a direct agent, a transitive middle with an implicitly reflexive reading is not routinely permitted unless the activity in question is also inherently or prototypically associated with self-interest. The sentence in (14) does not therefore reflect a regular "middle" use of T|Ko|ai:
(14)	*o ZuKpaxn^ xr|K£Tai tov Kr|pov 'Socrates gets the wax melted (for himself).'
Since such middles would naturally have had simple causative readings (= 'caused the wax to melt' etc.) that were virtually synonymous with those of their active equivalents, there would have been a strong motive either to discard them as redundant or to reinvent and revalidate them by assigning them distinctive meanings of their own. In this connection, consider the typical examples in (15):
(15)	alp« 'take'/a[pou|iai 'choose, anoSiS«|ii 'give back7anoSiSo|iai 'sell',
Ypa^u 'write'/Ypa^o^ai 'indict', nei0u 'persuade'/nei6o|ai 'obey', etc.
By contrast, transitive middles that were not assigned such "developed" meanings tended simply to drop out of use over time.
The relative infrequency of transitive middles with indirect reflexive readings (pace the standard grammars and lexica) explains why learners struggle to make sense of the vast majority of the middles they encounter in texts that obviously do not conform to the supposedly regular rule of interpretation. Equally, when learners look up a given transitive verb in a lexicon, they typically find that its middle in fact has a special sense, one that can only be connected with the supposedly "regular" indirect-reflexive sense via some tortuous special pleading of the type that tries to persuade us that "choose" is a semi-paraphrase of "take for oneself" etc. Pretending that these are somehow the straightforward middles of the corresponding actives in anything other than form is a disservice to students. They are clearly lexicalised verbs in their own right, with unpredi-cable meanings, and as such they deserve entries of their own in the lexicon.
The problems of the supposed "middle voice" do not end here, however. There are, for example, very large numbers of "middle only" (or deponent)
What's in the Middle? Two Voices or Three in Ancient Greek?
15
verbs that by definition do not enter into any voice alternation at all, cf. a few common examples in (16):
(16)	PoúAo|iai 'wish', -yryvo|iai 'become', o'fo|iai 'think', etc.
Nor should we forget the considerable numbers of paradigmatically "odd" middle forms, such as the inexplicable middle futures to otherwise normal active verbs, as in (17):
(17)	áKoú«/áKoúao|iai 'hear, |iav9áv«/|ia9r|ao|iai 'learn, ndaxw/neiao|ai 'suffer', etc.
At this point, we might very reasonably ask whether there really is a middle voice in Greek at all, given that it appears to be represented by a handful of lexically restricted implicit reflexives and a very large collection of oddities (viz. deponent verbs, middles with special meanings, and odd middle tenses for otherwise active verbs). In other words, it may be that the ancient grammarians basically got the middle right, at least from the general perspective that it cannot be reduced to any clear and simple definition and seems not to have any systematic relationship with the active or passive voices. On the face of it, then, it looks as if modern efforts to establish the credentials of the middle as a bona fide third voice are somewhat misconceived. My suspicion is that morphology, not for the first time, has taken precedence over syntax and semantics in the sense that the existence of marginally distinct middle morphology has been taken, incorrectly, to imply the existence of a functionally distinct middle voice (or diathesis).
3. VOICES IN PLATO REPUBLIC
The discussion above has involved a critical assessment of the standard proposition that a key property of the active-middle alternation is the regular addition of a secondary semantic role (patient or beneficiary) to an active agent, and that this "reflexivity" is marked by middle morphology. But this supposedly regular alternation appears to be far from regular in our corpus of Greek texts, where most middle forms are either "deponent" verbs with no active counterparts or show "irregular", i.e. semantically developed meanings vis-à-vis their corresponding actives (as suggested, the latter might very reasonably be added to the list of deponents as middle-only verbs in their own right).
So far, however, the argument has been based largely on theoretical considerations and assertions made without detailed numbers to support them.
16
Geoffrey Horrocks
To remedy this deficiency, book I of Plato's Republic was chosen as a reasonably "natural" example of dialogue among male members of the Athenian elite in the early 4th century BCE. First, every medio-passive verb form was collected (479 attested tokens) and assigned to the relevant lexical entry (167 different verbs, with an average frequency of 2.87, and with most falling in the range 1-5). Then the verbs were classified by type/function, with results as tabulated in (18):
(18) Verbs with middle-passive forms in the corpus
(a)	V with middle-passive morphology	167 of which:
(b)	V with middle-passive forms only (deponents-1)	75 V with "developed" middle sense (deponents-2)	40
(c)	V with passive sense (alternation ~ active)	40
(d)	V with a "regular" middle sense (alternation ~ active)	12
In (18a) we have the total number of verbs with medio-passive forms; in (b) the number of middle-only/deponent verbs and the number of verbs with middle forms that have semantically developed senses (which are in effect deponents too, as noted); in (c) the number of verbs that were clearly used as passives in alternation with actives; and in (d), the number of verbs that were used as middles in alternation with actives. (When a verb had the potential to be involved in a voice alternation that happened not to be attested in Republic 1, this was checked first in the Platonic corpus and then more widely, if necessary).
Of just 52 verbs that could in principle be involved in a regular voice alternation, 40 were deemed to be passive, and just 12 middle. Those middles with active equivalents of extremely rare or very late attestation (e.g., causative ano"y£uw beside ano"ysuo^ai, Pid(w beside PidZo^ai, evavtiw beside evavtiou^ai) were discounted.
Deponents proved to be by far the largest group (115 of 167 verbs). Importantly, some examples that might have in principle been taken as "regular" middles with active counterparts turned out to have middle forms that were consistently used with more abstract complements than their active counterparts and so showed a corresponding shift of meaning, however slight: e.g. dp^otto^ai 'tune (an instrument etc.)' vs. active 'fit/join', £vSsiKvu(o)^ai 'reveal (an opinion)' vs. active 'point out', npotiGs^ai 'propose (a theory)' vs. active 'place before/expose', ^statiGs^ai 'redefine (a word/concept)' vs. active 'place among/differently', SiopiZsaGai 'define (a word/concept)' vs. active 'divide/separate'. These were therefore counted as deponents. We might usefully compare here the famous example (19) from the beginning of the Republic:
What's in the Middle? Two Voices or Three in Ancient Greek?
17
(19)	.. .Kai |iou oniaGev o nai^ Xap6|ievoc xou l|iaxiou...
.and the slave boy, catching hold of my coat from behind.
Republic 327b
In the absence of any obvious reflexivity or self-interest, it seems that Plato here is using the middle of Xa|Pavw in the developed sense of 'grasp/take hold of, a usage that is in fact consistent throughout the corpus. It was perhaps initially modelled on anto|ai etc., involving contact with a part rather than seizure of the whole and therefore a genitive complement. Taken all together, this kind of evidence amply confirms the earlier suggestion that, by Plato's time, many middle paradigms, following a variety of models of development, had broken free from their active counterparts and become autonomous deponents with specialised meanings of their own.
There was also good evidence in the corpus that verbs with middle-only forms were still being created in Classical Greek, and that this tended to happen precisely when no clear semantic distinction between the active and middle had evolved. Consider the examples in (20):
(20)	(a) .. .aXXai; noXeu; snixeipeiv 8ouXoug8ai aSiK«^ ...
.to try to enslave other cities unjustly.
Republic 351b
(b) .. nepaiK^ PaaiXeia.xd^ sv xg r|neip« noXeu; e8ouX^ge .the Persian kingdom.. .enslaved the cities on the continent.
Thucydides 1.17.1
Any substantive difference between (20a) and (20b) is hard to detect, and any would-be explanatory references to reflexivity are not, in my view, convincing here. In Thucydides' time SouXw and SouXou|iai co-existed in free variation, but SouXou|iai turns out to be the sole survivor in Plato, and is consistently used as a middle-only verb by other authors of his period too, e.g. Demosthenes. A similar development is attested for the semantically related avSpanoSiio|iai. These data suggest that if the middle of a given verb failed to develop a distinctive meaning, one set of competing forms would eventually be dropped. A priori, we would expect this process to have favoured the active in most cases, and that is indeed generally the case. The opposite choice in the case of verbs of "enslavement" (and in other cases where the middle survives and it is the active that is dropped) presumably lies in the notion of advantage to the agent that is inherent in certain activities.
Turning now to the core cases of verbs with supposedly "regular" middles (just 12 out of 167 in the table in (18)), most seemed to be virtually
18
Geoffrey Horrocks
synonymous with their corresponding actives, with little suggestion of any "reflexivity" as a basis for distinguishing them. One might, of course, try to insist on a "regular" middle meaning simply because the grammars tell us it should be there, but this approach was not strongly supported by the contexts involved. Consider the representative examples given in (21)-(23), which are discussed individually below:
(21)	(a) . ..0| xu Zept^iw \ot8opou|ievu Kai Ai-yovxt oxi.
(Themistocles) who, when a man from Seriphus was reviling him and telling him that .
Republic 329e
(b) oukouv. . ,a'a8av6|ie8a.. .xtva.. AotSopouvxa Te auxov...; do we not. observe a man.. .reviling himself.?
Republic 440b
Can we honestly see the voice difference here as anything other than a matter of free choice? (Note too that reflexive meaning is carried by the active verb and an overt reflexive pronoun). There were several similar cases, including the commonly attested free variation between aKonw/aKonou^at.
Again, since something is provided for others rather than for the subject in both the examples in (22), any difference between them once more seems minimal:
(22)	(a) oukouv Kai «^eAiav £Kaaxr| xouxuv 'Siav xtva ^|iiv napexexat.;
and does not each of these (sc. arts) also provide us with a benefit that is peculiar to itself.?
Republic 346a
(b) .. .xouxo elvai, o naatv EKeivoi; x^v Suva|itv napeaxev «axe syyeveaGat... .this (sc. justice) is .what provided all those with the capacity to come into being.
Republic 433b
It may perhaps be that the middle emphasises provision as an inherent property of the provider or something similar (itself, in any case, an extended version of the reflexive theory), but there is, I think, a strong feeling of clutching at straws in trying to insist on any truly significant difference between this pair of sentences.
The same is evidently true of the pair in (23):
What's in the Middle? Two Voices or Three in Ancient Greek?
19
Tt8exat Se ye xou^ vo^ou^ eKaaxr| ^ apxn npo^ to auxg au^epov and each (form of) goverment enacts the laws with a view to its own advantage
Republic 338e
ouKouv snixeipouvxe^ vo^ou^ xt8evat xou^ |iev op0«^ xt8eaatv, xou^ Se xiva^ ouk op0«^;
in their attempts to enact laws do they (sc. rulers) not then enact some rightly and others not rightly?
Republic 339c
Specifically we might well ask why the first includes an overt expression of self-interest if the middle verb conveys this idea already? While it is perhaps still conceivable that the middle redundantly reinforces npoq to auxfl au^epov, it is hard once again to escape a feeling of special pleading if this particular path is followed.
It seems, then, that cases of virtually free variation are more common than is routinely acknowledged. At the same time, unequivocal cases of the supposedly prototypical middle use were actually very hard to find. The two best of the possible examples are those given in (24) and (25), where there does indeed seem to be a contrast involving the presence versus the absence of re-flexivity (though we should also compare (25) with (22) before jumping to this conclusion!):
(24)	(a) .. .ipavepui; nparro|ievot xfi; apxf<; eveKa |iia9ov...
.. .exacting pay openly for themselves in return for their service of rule
Republic 347b
(b) .. .npaTrovTOv Se o'l xa|iai xouxoiv xoiv 8eoiv...
.and the treasurers of these deities (sc. Hera and Zeus) shall exact (sc. the sum for the temple).
Laws774d
(25)	(a) Kai |i^v Kai opyava ye £Xwv napexea8ai uno nevia^...
and again, if from poverty he cannot provide himself with tools.
Republic 421d
(b) aAAa |ioi naXai npay^axa napexet.
but he has been creating issues for me for a long time
Phaedo 56e
(23) (a)
(b)
20
Geoffrey Horrocks
Nonetheless, such examples are exceedingly rare, not only in Plato but generally in Ancient Greek, and one might come closer to the truth, synchron-ically speaking, if one suggested that any implicit reflexivity in fact represents a very particular version of the familiar semantic specialisation process that was restricted to the middles of a small number of verbs with the right sort of meaning, as was suggested earlier (SiSaaKo^ai might be another), where self-interest or benefit to the subject is somehow a natural or inherent property of the activities in question.
4. CONCLUSION
The close analysis of a hopefully representative sample of Athenian prose tends strongly to confirm the preliminary conclusion that the alleged basis for an active-middle contrast, one that is routinely presented as the norm, is in fact anything but normal. It is in fact emphatically not the case that supposedly "regular" middles of potentially suitable verbs can be used productively to express either direct or indirect reflexivity. On the contrary, the few implicitly reflexive middles in the corpus studied here look more like one more case of semantic specialisation conditioned by lexical meaning. In any case, the overwhelming majority of the verbs with both active and medio-passive paradigms have clearly developed a sufficient degree of lexical and semantic dis-tinctiveness between their active and middle forms for the latter to be treated uncontroversially as autonomous deponent verbs.
Admittedly, this conclusion is based on the analysis of a small corpus taken from the work of only one author, and more research is obviously needed if the case for abandoning the middle as a true third voice is to be further substantiated. But it would be surprising if the preliminary indications from Republic I turned out to be freakishly misleading, and for now a strong prima facie case has been made that the putative middle voice in Ancient Greek really is a collection of disiecta membra, perhaps comprising some indirect reflections of a different kind of voice system originating in the prehistoric past. By the time Greek is first attested this earlier system had already been reinterpreted as a regular active-passive system, and the intractable residue of "middle" forms was either in the process of being lexicalised or of being progressively abandoned.
Geoffrey Horrocks Cambridge University gch1000@cam.ac.uk
What's in the Middle? Two Voices or Three in Ancient Greek?
21
BIBLIOGRAPHY
Allan, Rutger J. 2003. The Middle Voice in Ancient Greek: A Study in Polysemy. Leiden and Boston: Brill.
Allan, Rutger J. 2006. "Sophocles' Voice: Active, Middle and Passive in the Language of Sophocles." In Sophocles and the Greek Language: Aspects of Diction, Syntax and Pragmatics, edited by Irene J. F. de Jong and Albert Rijksbaron, 111-126. Leiden and Boston: Brill.
Allan, Rutger J. 2014. "Voice." In Encyclopedia of Ancient Greek Language and Linguistics Online, edited by Georgios K. Giannakis. Leiden and Boston: Brill. http://dx.doi. org/10.1163/2214-448X_eagll_COM_00000373.
Bakker, Egbert J. 1994. "Voice, Aspect and Aktionsart: Middle and Passive in Ancient Greek." In Barbara A. Fox and Paul J. Hopper, eds., 23-47.
Barber, Elizabeth J. W. 1975. "Voice - Beyond the Passive." In Proceedings of the First Annual Meeting of the Berkeley Linguistic Society, edited by Cathy Cogen, Henry Thompson, Graham Thurgood, Kenneth Whistler, and James Wright, 16-24. Berkeley: Berkeley Linguistic Society.
Chantraine, Pierre. 1927. "Le rôle des désinences moyennes en grec ancien." Revue de philologie, de littérature et d'histoire anciennes 5:153-165.
Conti, Luz. 2006. "Untersuchung der sogenannten inhärent reziproken Verben im Altgriechischen: Semantische und syntaktische Eigenschaften." Historische Sprachforschung 119:168-185.
Duhoux, Yves. 2000. Le verbe grec ancien: éléments de morphologie et de syntaxe historiques. Deuxième édition, revue et augmentée. Louvain-la-Neuve: Peeters.
Fox, Barbara A., and Paul J. Hopper, eds. 1994. Voice: Form and Function. Amsterdam and Philadelphia: John Benjamins.
Garcia Gual, Carlos. 1970. El sistema diatético en el verbo griego. Madrid: Instituto Antonio de Nebrija.
Gildersleeve, Basil L. 1900-1911. Syntax of Classical Greek: From Homer to Demosthenes. 2 vols. New York: American Book Company.
Kemmer, Suzanne. 1993. The Middle Voice. Amsterdam and Philadelphia: John Benjamins.
Kühner, Raphael, and Bernhard Gerth. 1889-1904. Ausführliche Grammatik der griechischen Sprache. 2. Teil, Satzlehre. 2 vols. Hannover: Hahnsche Hofbuchhandlung.
Prévot, André. 1935. L'aoriste grec en -then. Paris: Champion.
Risselada, Rodie. 1987. "Voice in Ancient Greek: Reflexives and Passives." In Ins and Outs of the Predication, edited by Johan van der Auwera and Louis Goossens, 123-136. Dordrecht: Foris Publications.
Rijksbaron, Albert. 2006. The Syntax and Semantics of the Verb in Classical Greek: An Introduction. 3rd edition. Chicago: Chicago University Press.
Rijksbaron, Albert. 2018. "The Treatment of the Greek Middle Voice by the Ancient Grammarians." In Albert Rijksbaron, Form and Function in Greek Grammar: Linguistic Contributions to the Study of Greek Literature, Amsterdam Studies in Classical Philology 30, edited by Rutger Allan, Evert van Emde Boas, and Luuk Huitink, 357-369. Leiden: Brill.
Ruipérez, Martin S. 1988. "Sur la structure des oppositions de voix dans le verbe grec." In In the Footsteps of Raphael Kühner: Proceedings of the International Colloquium in Commemoration of the 150th Anniversary of the Publication of Raphael Kühner's
22
Geoffrey Horrocks
'Ausführliche Grammatik der griechischen Sprache, II. Theil: Syntaxe', edited by Albert Rijksbaron, Hotze A. Mulder, and Gerry C. Wakker, 255-264. Amsterdam: Gieben.
Schwyzer, Eduard, and Albert Debrunner. 1959. Griechische Grammatik. 2 vols. Munich: C. H. Beck.
Smyth, Herbert W. 1956. Greek Grammar. Revised by Gordon M. Messing. Cambridge,
MA: Harvard University Press. van Emde Boas, Evert, Albert Rijksbaron, Luuk Huitink, and Mathieu de Bakker. 2019. The Cambridge Grammar of Classical Greek. Cambridge: Cambridge University Press.
Willi, Andreas. 2018. Origins of the Greek Verb. Cambridge and New York: Cambridge University Press.
ABSTRACT
It has long been taken for granted in reference works, grammars and elementary introductions that Ancient Greek had three grammatical voices, active, passive and middle. Yet scholars have always had great difficulty in characterising the middle voice in a straightforward and convincing way, and language learners are often perplexed to find that most of the middles they find in texts fail to exemplify the function, usually involving some notion of self interest, that is typically ascribed to this voice. This article therefore re-examines the Ancient Greek middle, both through the lens of a general survey of "middle voice" functions across languages, and through the analysis of all the medio-passive verb forms attested in Book 1 of Plato's Republic.
The principal observations are that Ancient Greek middles do not represent a regular pattern of usage either from a typological point of view or as employed specifically in Republic 1 (the database is in fact partly extended to other works). Accordingly, the main conclusion is that the Ancient Greek middle is not a grammatical voice sensu stricto, i.e. a regular syntactic alternation applying to all verbs with a given set of properties and expressed by a regular morphological form with a predictable semantic function. Rather, it appears to be a convenient collective name for a large set of "autonomous" verb forms that are either clearly deponent (i.e., have no active counterparts) or that have been lexi-calised in a specialised meaning vis-à-vis their supposed active counterparts (i.e., are also deponents in practice, despite appearances). In all probability, therefore, medio-passive morphology, whatever it once represented in terms of function, was recharacterised pre-historically as "passive" morphology, leaving a residue of verbs exhibiting forms with nonpassive functions. Presumably, these survived as "middles" only because they had no active counterparts or had been assigned innovative meanings that distinguished them from any formally related actives.
Keywords: active voice, middle voice, passive voice, deponent verb, semantic specialisation
What's in the Middle? Two Voices or Three in Ancient Greek?
23
POVZETEK
Kaj je na sredini? Dva ali trije načini v stari grščini?
Referenčna dela, slovnice in najelementarnejši jezikovni uvodi po tradiciji kot samo po sebi umevno jemljejo dejstvo, da je imela stara grščina tri načine, aktiv, pasiv in medij (ali »srednjik«). A filologi se vsakič znova znajdejo v hudi zadregi, ko je treba medij jasno in prepričljivo opredeliti, medtem ko študentje stare grščine pogosto presenečeni opazijo, da večina oblik medija v izvirnih besedilih ne ustreza vlogi, ki se mu običajno pripisuje in za katero naj bi bila značilna določena mera subjektovega osebnega interesa. Pričujoči prispevek torej na novo odpira vprašanje starogrškega medija, in sicer z vidika tipološkega pregleda »medijalnih« funkcij, vključuje pa tudi analizo mediopasnih glagolskih oblik, izpričanih v 1. knjigi Platonove Države.
Poglavitne ugotovitve kažejo, da niti v tipološkem smislu niti z vidika 1. knjige Platonove Države (korpus je v resnici nekoliko širši in vsebuje tudi odlomke drugih del) raba medija ne sledi jasnemu vzorcu. Iz tega izhaja najpomembnejši zaključek prispevka, da namreč starogrški medij ni glagolski način v pravem pomenu besede in da torej ne moremo govoriti o pravilni skladenjski tvorbi, ki bi se uporabljala v primeru vseh glagolskih oblik z določenimi lastnostmi in se izražala s pravilnimi oblikoslovnimi sredstvi s predvidljivo semantično funkcijo. Nasprotno, izkaže se, da gre za prikladno kolektivno ime za veliko skupino »avtonomnih« glagolskih oblik, ki so bodisi očitno deponentne (t.j. nimajo aktivnih ustreznic) ali pa so bile, v nasprotju s hipotetičnimi aktivnimi ustreznicami, leksikalizirane za izražanje specializiranih pomenov. Po vsej verjetnosti se je torej mediopasivno oblikoslovje, četudi je morda nekoč predstavljalo posebno funkcijo, v predzgodovinski dobi rein-terpretiralo kot »pasivno«, pri čemer so se kot okameneli ostanki ohranile glagolske oblike z nepasivnimi funkcijami. Domnevati smemo, da se so slednje ohranile kot »medijalne« zgolj zato, ker niso imele aktivnih vzporednic ali ker so pridobile drugotne pomene, po katerih so se razlikovale od aktivnih, v formalnem pogledu z njimi povezanih oblik.
Ključne besede: aktiv, medij, pasiv, deponentnik, pomenska specializacija