As presented in section 2, passive inflection includes the -at/-et/-tat/-tet causative suffix. However, it is often unclear, and must have been unclear for Old and Middle Hungarian speakers, as well, whether -(t)At is part of the passive inflection, or – as a derivational suffix – it is part of the base verb. The source of the problem is that the -(t)At suffix is ambiguous (see Komlósy 1994; 2000; Horvath & Siloni 2011). When combining with agentive (transitive and unergative) verbs, it functions as a causative (or, in the terminology of Komlósy 2000, factitive) suffix, introducing an agent that manipulates the original agent of the base verb to do the action denoted by the verb, as happens in the case of dolgoz-tat ‘make someone work’ or nevet-tet ‘make someone laugh’. But -(t)At can also function as a transitivizing suffix, creating a transitive verb from an unaccusative, as happens in the case of forg-at ‘turn’, derived from the unaccusative forog, mozg-at ‘move’, derived from the unaccusative mozog, or változ-tat ‘change’, derived from the unaccusative változ-ik. In structural terms: -(t)At can be a v head merging with a V, yielding a transitive verb as in forg-at ‘turn’, and it can be a causative head merging with a v, yielding a causative verb, as in forgat-tat ‘make turn’. The causative – transitive distinction, however, can be blurred if the causative -(t)At combines with an unergative verb or a transitive verb with an implicit theme, and the -(t)At-marked verb denotes a regular transitive activity involving no distinct, incidental momentum of causation. Such causative verbs can easily be reanalyzed as simple transitive verbs with the agent of the base verb interpreted as a patient – see, for example, e-tet ‘feed’, i-tat ‘make/let drink’, al-tat ‘put to sleep’, hív-at ‘call, send for someone’, éhez-tet ‘let starve’, ül-tet ‘seat, plant’, ismer-tet ‘make known, familiarize’, kopog-tat ‘knock’, ütköz-tet ‘clash’, öltöz-tet ‘dress’, legel-tet ‘make/let graze’, siet-tet ‘make hurry, urge’, figyelmez-tet ‘make pay attention, warn’, etc. When interpreting these verbs, the intuition is – and must have been – uncertain whether the -(t)At suffix is the causative component of passive inflection, or it is a transitivizer and has to be spelled out again in passive sentences, i.e., whether ‘he was dressed’ corresponds to (ő) öltöz-tet-ett or öltöztet-tet-ett, or whether éhez-tet-ett is an active transitive verb with a pro subject and a pro object, or a passive verb with a pro subject, i.e.:

In some cases, Old and Middle Hungarian speakers eliminated this problem by doubling the -(t)At suffix on passive verbs. The 150 passive verbs I selected from the Gospels of Müncheni Codex do not contain any doubling yet, however, in the subsequent translations the proportion of passive verbs with -(t)At doubling is fairly high:

The passive verbs displaying -(t)At doubling in the above corpus include: hív-at-tat ‘call-pass’, vág-at-tat ‘cut-pass’, elvét-et-tet ‘away.take-pass’, ismer-tet-tet ‘familiarize-pass’, kárhoz-tat-tat ‘condemn-pass’, gyötr-et-tet ’torture-pass’, hagy-at-tat ‘leave-pass’, hoz-at-tat ‘bring-pass’, kivet-tet-tet ‘eject-pass’, hány-at-tat ‘afflict-pass’, ad-at-tat ‘give-pass’, tör-et-tet ‘break-pass’, nyom-at-tat ‘press-pass’, ver-et-tet ‘beat-pass’, sóz-at-tat ‘salt-pass’, fog-at-tat ‘catch-pass’, mond-at-tat ‘say-pass’. One may surmise that these are passivized causatives verb forms, with the two -(t)At suffixes having distinct, causativizing and passivizing functions. In fact, the contexts exclude the passivized causative readings; they support the claim that these are simple passive verbs, with the first -(t)At (mis)analyzed as part of the stem, as illustrated in (13a–c):

The uncertain status of verb-final -(t)At is also evident in Hungarian grammars from the Middle Hungarian period. Nova Grammatica Ungarica by Albert Szenczi Molnár (1610) presents the passive paradigm of the verb kér ‘ask for’ in two alternative versions: with a single (kér-et-ik) and with a doubled -(t)At suffix (kér-et-tet-ik). The discussion of the passive conjugation in the Hungarian grammar of György Komáromi Csipkés (1655: 108) also contains verb forms both with and without causative doubling (ver-et-em and ver-et-tet-em ‘(I) am beaten’, but Komáromi Csipkés – alone among all grammarians – regards the version with doubling a passivized causative. The so-called Debreceni grammatika presents two passive paradigms, those of the verbs zár ‘close’ and vesz ‘take’. In the paradigms of zár, the present and past forms contain a single causative morpheme (zár-at-ik, zár-at-ott), whereas the conditional and the subjunctive forms contain causative doubling: zár-at-tat-nám, zár-at-tas-sam. The passive infinitive is listed in two versions: zár-at-ni and zár-at-tat-ni. In Kolozsvári grammatika from the early 18th century, the list of passive verbs on p. 27v contains, in addition to such regular forms as szeret-tet-el ‘(you_sg) are being loved’, tanít-tat-ol ‘(you_sg) are being taught’, olvas-tat-ol ‘(you_sg) are being read’, the doubled hall-at-tat-ott ‘(he) was heard’ (Lőrinczi 1998: 52). In Sámuel Fejérvári’s Institutiones, written between 1755–1773, p. 90, we find both szeret-tet-em ‘I am loved’, olvas-tat-om ‘I am read’, mutskol-tat-om ‘I am dishonored’ and the doubled ken-et-tet-em ‘I am anointed’ (Lőrinczi 1998: 195). Verseghy’s grammar (1805) incriminates the use of double -(t)At.

The opacity of -(t)At-marked verb forms was aggravated by a further factor: the incompleteness and attrition of the -ik paradigms of subject-verb agreement. The passive verb takes an agreement morpheme of the -ik conjugation, the unaccusative conjugation of Old Hungarian. Most suffixes of this conjugation are obsolete in present-day Hungarian; they have been replaced by the corresponding suffixes of the default indefinite conjugation.10 In Old Hungarian, when the -ik paradigms were still more complete, they differed from the default indefinite paradigms in present singular 1st, 2nd, and 3rd person, in conditional 1st and 3rd person, and in subjunctive 1st and 3rd person. In the paradigms of the verb lak-ik ‘stay, live (somewhere)’, shown in (14), the verb forms that bear morphemes different from the corresponding regular indefinite agreement suffixes are spelled in boldface:

Where the morpheme is identical with that of the indefinite paradigm, the verb form is ambiguous between a passive and an active causative interpretation. Since Hungarian has not only subject pro-drop but also object pro-drop, these homophonies can lead to ambiguous sentences – at least when both the agent and the theme of the verb are human. For example:

The passive verb forms in the English translations of (15b), (16b), and (17b) are misleading; the Hungarian sentences are active, containing accusative-marked objects. The lack of passive morphology on the verb is evident in examples with a 3SG subject, where the verb lacks the 3SG -ik suffix of 3SG present passive verbs; it bears a suffix cross-referencing both a 3SG subject and a definite object:

Some unaccusative morphemes different from the corresponding morphemes of the indefinite paradigm happen to be homophonous with verbal agreement morphemes of other paradigms. Namely:

These homophonies can lead to further ambiguities. In (20), (21), and (22), the same verb forms are interpreted passively in the (a) examples, and actively in the (b) examples.

In the Middle Hungarian period, the incomplete -ik conjugation started to disappear, i.e., more and more of its suffixes were replaced by suffixes of the default indefinite paradigms. At present, the -ik conjugation differs from the regular indefinite conjugation in 1SG and 3SG present tense in Standard Hungarian, and only in 3SG present tense in substandard varieties. The attrition of the -ik paradigms was already evident in the 18th century (though there were efforts by Révai 1778 to resurrect it). Verseghy’s Hungarian grammar (1805) says it is used only by a part of Hungarians, and it is used tentatively even by them. Simonyi’s prescriptive school grammar (1879) already characterizes the use of the -ik paradigm as optional.

In sum: Neither element of passive inflection was fully transparent even in the Old Hungarian period. Its causative component, the -(t)At suffix, was formally identical with the morpheme deriving the transitive alternant in many transitive – unaccusative verb pairs, which led to ambiguities in the case of pro-drop. The -(t)At suffix assumes passive meaning when combined with an agreement morpheme of the -ik conjugation. But the -ik paradigms were deficient in the Old Hungarian period already, and their attrition accelerated in the Middle Hungarian period, with more and more agreement suffixes replaced by the corresponding morphemes of the indefinite conjugation. This was the source of further ambiguities.

Middle Hungarian speakers tried to disambiguate passive verb forms by -(t)At doubling, but – as will be demonstrated below – eventually it proved to be more economical to give up the passive conjugation altogether and to use alternative constructions with similar expressive power.

The passive voice, encumbered by the processing problems discussed above, was not indispensable in Old Hungarian any more. In SOV Proto-Hungarian, it presumably served to topicalize the internal argument, to focus the external argument, or to eliminate the external argument. In Old Hungarian, the generalization of object marking to all objects, and the evolution of pre-VP focus and topic positions accessible to subjects and objects alike made the topicalization of the internal argument and the focusing of the external argument possible without the use of the passive construction – as shown by examples from the earliest surviving Hungarian documents:

The replacement of the passive construction by object topicalization and subject focusing can be traced in subsequent Bible translations:

Old Hungarian also had means to express external argument demotion. Interestingly, these means are listed in Debreceni grammatika (1795) already (for a recent review, see Szily 1999).

(i) When the subject is unknown, or its identity is irrelevant, often a 3Pl verb is used with a pro subject – see (24b) and (26):

(ii) Another possibility is to supply the transitive verb with the -ód(ik)/-őd(ik) anticausative suffix (Halm 2020: 4), as in (27a), or to use its unaccusative alternant, as in (27b), containing épül, the unaccusative pair of épít ‘build’ (27b):

(iii) A third possibility is to denote the result state of the patient of the event with a -vA participle phrase (in a different terminology, converb phrase) + van ‘be’ construction:

(iv) In later translations, more and more passive forms are replaced by idiosyncratic solutions, for example, by nominal or adjectival predicates.

The gradual replacement of the passive constructions by these means is also evident in the Bible translations, e.g.:

The distribution of these replacement strategies among the non-passive equivalents of the 150 passive constructions of Müncheni Codex in later translations is not uniform:

The distribution of the replacement solutions depends on their productivity. The use of a 3rd person plural verb with a pro subject, the so-called impersonal construction, is the most common strategy presumably because it can be used with any verb. The -Ódik anticausative suffix is also fairly productive. It can combine with most transitive verbs, however, if the lexicon already contains an intransitive version of the given verb, it often blocks the use of the anticausative form. For example, the transitive verbs főz ‘cook’ and süt ‘bake’ form transitive – unaccusative pairs with fő and sül, and the latter ban the anticausative forms főz-ődik and süt-ődik (cf. Halm 2020: 13). If a transitive verb, for example, nyit ‘open’, has two intransitive equivalents: nyit-ód-ik and nyíl-ik, the anticausative nyitódik sounds more marked in most contexts, except when e.g. the sentence refers to a non-operating agent or force: Nem nyitódik a bal hátsó ajtóm ‘My left rear door won’t open’.

The use of the -vA participle phrase + van construction is limited both functionally and lexically. It serves to denote the result state of the theme argument. Its use is restricted to verb classes with particular aspectual and semantic features, and the restrictions are related to the result state function of the construction in ill-understood ways. Bartos (2009) provides the following non-exhaustive list of constraints:

In sum: in the present-day Ugric languages and presumably in Proto-Hungarian, too, the primary function of passivization was to topicalize the internal argument, and to focus or to demote the external argument. In the Old Hungarian clause structure, topicalization and focusing could already target any argument without passivization. There were also regular means of agent demotion: the agent could be represented by a non-referring, existentially bound pro subject; it could be eliminated by the use of an anticausative or unaccusative verb; and it was also absent or demoted to a PP in the -vA participle phrase + be-verb construction predicating the result state of the internal argument.

The structure of the Hungarian passive construction is represented along the lines of Bruening’s (2013) proposal. It involves a Passive head, realized by the -tAt causative morpheme. When -tAt functions as Passive, it does not project a specifier and does not introduce a causer argument. The complement of Passive is a vP (alternatively labeled VoiceP) that introduces an external argument without projecting it. Its agent or force theta role can be expressed by a PP adjoined to vP. The sentence in (34a), for example, is assigned structure (34b):

Tense and agreement are assumed to project a joint IP. Passive voice, and tense + agreement are bound morphemes picked up by the verb via head movement. As claimed by Piñón (1992) and Surányi (2009), Spec,IP is filled by the verbal particle or some other predicative element forming a complex predicate with the verb. The internal argument gets nominative case in situ.

The impersonal construction, the most widely used replacement strategy, involves a vP whose external argument is represented by an existentially bound, phonologically empty impersonal pro. This pro_imp has a [human] feature; thus in (35a), for example, it cannot refer to winds or dogs:

The -Ód- anticausative suffix is argued by Halm (2020) to realize Middle voice in Hungarian. Like the Passive head,, it does not project an external argument, and it is complemented by a vP whose external argument is not expressed, either:

The non-projected external argument of the v usually either cannot – or only marginally can –be expressed by a PP, unlike in the passive construction:

The ungrammaticality or marginality of these examples may be related to the fact that the -Ód- suffix – as opposed to the -(t)At- morpheme – is not fully productive. Therefore, while the passive -(t)At is clearly an inflectional suffix combining with a syntactically visible vP, -Ód- may be for most speakers a derivational suffix creating an unaccusative lexical verb whose inner components are inactive in syntax. These speakers presumably assimilate anticausative verbs to the other types of unaccusatives, and identify them as Vs, as represented in (38b). In this structure, no agent or force argument is introduced on any level of representation.

As shown above, the lack of a phonologically salient external argument in the passive, impersonal, and unaccusative constructions is the consequence of different syntactic structures. The external argument of the impersonal construction is present in syntax; the external argument of the passive is not projected syntactically, but the agent is part of the argument structure semantically; whereas in the the unaccusative construction the external argument is absent in every respect. The presence of an agent-oriented adverb presupposes the presence of an agent in semantic structure. Consequently, it is licensed in impersonal and passive sentences (39a, b), but is anomalous in unaccusative constructions (39c):

The external argument of an impersonal sentence can be the subject of depictive secondary predication (40a). The non-projected external argument of a passive construction, which is nevertheless present in the argument structure of the verb, can license a depictive adverb if it can be understood to denote the manner of the agent’s action (40b). Unaccusative sentences do not allow a depictive predicated about the external argument (40c).

Obligatory control is a structural relation in which the external argument of an impersonal verb can, the syntactically unrealized agent of a passive verb cannot, take part (41a,b). The missing external argument of an unaccusative verb naturally cannot control; in the marginally possible reading of (41c), the controller is the theme, and the verb akar ‘want’ is coerced to mean ‘be about’.

The external argument of an impersonal verb can bind an anaphor (42a, a’). The syntactically unrealized external argument of a passive verb cannot bind the internal argument promoted to subject. Interestingly, if the anaphor is a non-subject, ungrammaticality is less sharp (42b’). The non-existent external argument of an unaccusative verb naturally cannot bind (42c).

The least common, most restricted passive replacement pattern, the result state construction involves a -vA participle phrase and the verb van ‘be’. The -vA participial suffix can be merged with either an agentless transitive verb (43a) or an unaccusative verb (43b).12 In the former case, the agent theta role can be expressed by a PP adjunct.

In traditional grammars, stative -vA participles are labelled adverbial participles because they express states, and because – like adverbs and PPs, and unlike verbal, nominal and adjectival predicates – they cannot function as predicates of finite clauses on their own; they require the presence of a be-verb. Recent approaches, e.g., Dékány & Hegedűs (2021), do not assume a lexical category ‘adverb’; lexical items that have traditionally been categorized as adverbs are viewed by them as PPs headed by an opaque P head. In the case of the suffix -vA, the opaque P is the lative -A suffix, which is also present in the illative -bA, sublative -rA and translative -vÁ suffixes (A. Jászó 1991: 344–345). The -v of -vA can be traced back to -m-, the suffix of the perfect participle and nomen actionis in Proto-Ugric and in the present-day Ob-Ugric languages.13 It evolved into -v- via -β-. Participle phrases are assumed to be extended verbal projections subsumed by an AdjP layer (e.g. Lundquist 2013), hence -vA-marked participle phrases are historically AdjPs combining with a lative suffix.

From a synchronic point of view, the -vA participial suffix is a manifestation of [-Tense] inflection, as represented in (44); the lative-resultative component is part of its feature specification. When the -vA participle phrase is built on a transitive verb, the vP is passivized; that is, it complements a morphologically null Pass head that projects no specifier. The external argument of the vP is not expressed, but its agent or force theta role can optionally be realized as a PP.

The theme argument gets nominative case in the matrix clause, where the impersonal finite be-verb has no subject. It tends to be raised to the topic position – see (45a). The verbal particle moves into the specifier of the matrix IP for prosodic reasons (Szendrői 2004). In (45b), the PP is also raised into the matrix clause; it occupies the focus position.

The spellout of the agent as a PP is restricted in various ways (Bartos 2009). The restrictions appear to be related to the result state denoting function of the participle phrase. In (46b), the lack of the telicizing verbal particle renders the sentence unacceptable:

When the base verb of the resultative -vA phrase is unaccusative, no vP is projected:

The theme argument undergoes raising to the matrix clause in the case of (47), too. Since no vP is projected, no agent PP is possible:

Though the agent is syntactically unexpressed in structures of type (44), it can license an agent-oriented adverb (49a). Such adverbs cannot appear if the base verb is unaccusative (49b), i.e., if no vP is projected:

A -vA phrase built on a vP not projecting a specifier can allow an agent-oriented depictive if it can be understood as the manner of the agent’s action, as is the case with gondatlanul. In the absence of a vP projection, the depictive is not possible.

The presence of vP in (44) as opposed to (47) has various further consequences. As pointed out by Bartos (2009), circumstantial adverbs can only appear if a vP is projected:

vA-phrases derived from transitive verbs, unlike those derived from unaccusatives, can also combine with the perfective copula lesz ‘become’, in which case they denote inchoative result states, i.e., result states including their starting point. Apparently, the starting point of a state cannot be denoted without implying an agent, even if it is projected without being expressed:

In sum: In the passive and the passive-replacing constructions, the external argument is demoted to different degrees, with differing syntactic consequences. The subject of the impersonal construction is only referentially deficient. It is fully active syntactically, capable of controlling, binding, and licensing agent-oriented adverbs and depictives. The external arguments of the passive construction and of the passive vP of the resultative -vA participle phrase are not projected in syntax, but they are part of the argument structure of the v. Therefore, they cannot participate in syntactic relations interpreted on the clausal spine such as control and binding, but they may participate in semantically based relations, licensing agent-related adverbs and depictives denoting the manner of the agent’s action. Finite clauses and -vA participle phrases involving an anticausative or some other type of unaccusative lexical head are agentless in every respect, hence they cannot contain agent-related elements.