Aspectual cognate objects in Hungarian

2.1 Thematic direct objects

The Hungarian sentence is divided into a topic part and a predicate part. As an action or state can be predicated about any of its participants, either the grammatical subject or the grammatical object can occupy the position associated with the topic function. In the former case, as shown below, the accusative object remains in the predicate part and appears postverbally in the neutral sentence (i.e., in a declarative without progressive aspect, negation or narrow focus):

The preverbal section of the predicate phrase contains operator positions, which can be occupied, for instance, by a focus constituent expressing exhaustive identification. Crucially, if the preverbal operator field contains such an element, the non-neutral sentence triggers the postverbal position of the verbal particle:

The question of an element being in a preverbal versus postverbal position is an essential feature of the Hungarian sentence: whereas the preverbal position is reserved for non-referential (that is, predicative) elements such as verbal particles, resultative and goal-denoting predicates or bare nouns, the postverbal (argument) position is reserved for referential elements. As such, Hungarian (singular and plural) bare nominals immediately precede the finite verb in neutral sentences while they are postverbal in non-neutral sentences. That they cannot stand postverbally (in an argument position) in a neutral sentence but can only survive in a preverbal position as an aspectual operator, a focus or a distributive quantifier (with is ʻalso’) is shown in the following pair of sentences:

Interestingly, (5a) improves in grammaticality if a preverbal or verbal focus is introduced, as shown below:

In such a case the postverbal bare nominal can be interpreted as a focus in situ, given that a preverbal focus also licenses postverbal foci (É. Kiss 2002: 29).

In addition, the Hungarian verb agrees not only with the subject but also with the direct object. In a nutshell, whereas a definite object triggers the definite conjugation on the verb (see the verb form táncolta in [7a]), an indefinite object triggers the indefinite conjugation on the same verb (see the verb form táncolt in [7b]). As shown below, the same indefinite agreement marker appears on the verb when it is used either with a bare noun (singular or plural) or in an intransitive configuration:

As for their syntax, Hungarian bare nominals have been shown to be base-generated postverbally in VP but, as non-referential elements, they are preposed from this phrase into a position in the preverbal section of the predicate, more precisely into the specifier of a functional phrase. This is either aspect phrase (AspP), as in É. Kiss (2002), or predicate phrase (PredP), subsumed by AspP, among others, as in É. Kiss (2008). Furthermore, they trigger verb raising to Asp or Pred, respectively (whereas in the former account the bare noun is raised to the specifier of AspP and the verb in Asp eventually merges with the bare nominal, in the latter account the functional phrase to which the bare noun moves cannot be AspP as demonstrated by the lack of the aspectual operator role of these nominals as a bare nominal–V complex is usually interpreted as atelic). As bare object nominals are predicative elements just like verbal particles and result or goal-denoting predicates, and the (above) specifier position can only host one element, they are mutually exclusive – at least in neutral sentences – and are subject to severe co-occurrence restrictions.

2.2 Cognate objects

In a COC a verb takes an object expressed by a DP, the head noun of which is a nominalization of the verb stem. In other words, the CO is a semantic and morphological ‘copy’ of the verb. According to their most recent classification (Horrocks and Stavrou 2010; Lavidas 2013a, 2013b, 2014, 2018), these constructions can be transitive, transitivizing or aspectual.

Transitive COCs are built on transitive verbs, which select a subject and obligatorily an object as well. This object can be either a regular direct object or a cognate object. In addition, these objects have a concrete meaning, they can be singular or plural, and they can freely be passivized. Two relevant Hungarian examples are:

Similarly to the Biblical texts of Hebrew (Horrocks and Stavrou 2010; Mittwoch 1998; Pereltsvaig 2002), Greek (Gianollo and Lavidas 2013) or Early Modern English (Lavidas 2018), Biblical Hungarian has a wide variety of transitive (and, as shown below, transitivizing) COCs, which are frequently associated with a stylistic choice and are claimed to have an emphatic function. They reflect the influence of the source texts, as they copy the pleonastic structures occurring in the corresponding Greek or Hebrew verses (see [8a]). Furthermore, in everyday speech the CO can exhibit all the syntactic properties of a non-cognate thematic object (see [8b]).

Transitivizing COCs are built on so-called derived unergative verbs, which take a subject and, optionally, an object as well, which can be a cognate or a non-cognate object (e.g. synonym or hyponym). As argued in Mittwoch (1998), Pereltsvaig (2002), de Swart (2007), Horrocks and Stavrou (2010) or Lavidas (2013a), in the former case the objects are only accidentally morphological COs of the verb: they do not denote events, they are fully referential arguments, hence they do not share the properties of ACOs. I give two canonical examples built on verbs of (re-)creation taken from Biblical Hungarian (9a) and everyday speech (9b):

Again, the CO has all the syntactic properties of a non-cognate thematic object: in (9b) it appears as a premodified (NP/NumP) bare noun occupying the preverbal position and triggering the indefinite conjugation of the verb (cf. the verb forms rajzolt versus rajzolta).

Aspectual COCs are built on so-called prototypical unergative verbs, which require only a subject and, exceptionally, also an accusative nominal, which is exclusively a cognate object. The most important features of these constructions and these objects are the following: the CO is not a referential object and cannot be treated as a subcategorized (internal) argument, hence it fails the canonical tests that apply to referential objects; it cannot be replaced by a similar noun (e.g. synonym or hyponym); and the main function of the entire construction is the expression of a limited event with beginning and end (hence the term ‘aspectual’). We may find such constructions in diverse translations of the Holy Bible (10a), where, again, they reflect the translator’s fidelity to the source text, but not in everyday speech (10b):

As for their syntax, Hungarian objects that are literally cognate with the (transitive or derived unergative) verb they accompany can be claimed to share the properties of non-cognate accusative nominals: bare cognate nouns are base-generated in VP but move to [Spec, AspP]/[Spec, PredP] and trigger V-to-Asp/Pred movement. Moreover, when the CO is preverbally expressed by a bare nominal, its co-occurrence with another predicative element such as a verbal particle or a result/goal-denoting predicate is excluded since they would compete for the same syntactic position.

In sum, this language has transitive and transitivizing COCs both in Biblical texts and in everyday speech. But crucially, with the exception of some Biblical examples, in Hungarian there are no ACOCs in the sense defined in Horrocks and Stavrou (2010) or Lavidas (2013a, 2013b, 2014, 2018. Instead, as argued in this paper, Hungarian has POs that fulfil the function of the ACO in the language.

One important caveat is in order here: in this paper I only analyse the subclass of POs I have briefly presented above. However, there is another subclass of pseudo-objects, found in resultative constructions but not discussed in this paper, which includes reflexives such as magát ‘himself.acc’ in álomba sírta magát ‘(he) cried himself to sleep’ and nominals denoting body parts such as a szemét ‘his eyes.acc’ in kisírta a szemét ‘(he) cried his eyes out’. For more details on these and similar pseudo-objects in Hungarian, see Csirmaz (2008).

2.3 Pseudo-objects

The PO egyet ‘one.acc’ is formally the accusative-marked (-t) numeral or indefinite article egy ‘one/a(n)’, with -e- being the link vowel on the stem ending with a consonant, in accordance with the vowel harmony characteristic of this language.^[5] In standard, present-day Hungarian, it follows the intransitive activity verb and it is compatible with a wide variety of intransitive(ly used) verbs:^{[6] , [7]}

POs of class (b) and (c), also called accusative adjectives (Csirmaz 2008), are built from adjectives (such as jó ‘good’ or harsány ‘uproarious’) and not adverbs (such as jól ‘well’ or harsányan ‘uproariously’, derived from the corresponding adjectives by means of a manner adverbial suffix), and are literally not cognate with the prototypical unergative verb they accompany. They can take intensifiers as premodifiers (12a), and comparative (12b) or superlative morphemes (12c):

As expected, in the last case the definite object triggers the definite conjugation on the verb, as shown by the verb forms kacagott in (12a) and (12b) but kacagta in (12c).

In addition, they can be supplied with a plural marker -k (and an additional linking vowel), which makes them similar to elliptic adjectives (rather than adverbs), where, in the absence of the head noun, the plural and accusative markers are attached directly to the adjective.^[8]

While POs of class (b) are compatible with a wide variety of intransitive(ly used) verbs, POs of class (c) may impose severe s-selectional restrictions on the verb they accompany, as shown by the infelicitous VP below. This property is related to the adverbial interpretation of these latter POs (see Section 5.3).

Whereas singular POs accompanied by egy ‘one/a(n)’ have the distribution of indefinite singular objects as they can appear either preverbally or postverbally (cf. [15a]), singular (and plural) POs not accompanied by egy ‘one/a(n)’ mirror the position and distribution of Hungarian bare nominals as they immediately precede the finite verb in neutral sentences (cf. [15b]); see also Kiefer (2006):

More precisely, POs of class (b) and (c) may either appear preverbally or postverbally, and in both cases they can occur either as an indefinite or as a bare noun but in the latter case they must occur in a non-neutral sentence. As such, similarly to a preverbal or verbal focus licensing a thematic bare nominal postverbally (see [6] above), such a focus can also license a (singular or plural) PO in the same postverbal position, as in (16):

Irrespective of their position, POs – as indefinite or (singular/plural) bare nominals – trigger the indefinite conjugation of the verb. This is strongly connected to the indefinite restriction on ACOs more generally, which states that these objects are generally indefinite (see Section 5).

The following examples illustrate the co-occurrence restrictions of POs with a subcategorized (non-cognate or cognate) object (17a), a verbal particle (17b) or a stative verb (17c).

With respect to their syntax, Csirmaz (2008) argues that when the PO accompanies an (unergative) activity verb, it functions as an argument, it takes the vacant direct object position and merges inside the VP. However, when it modifies an (unaccusative) semelfactive verb, it functions as an adjunct and must adjoin PredP. In addition, it moves to [Spec, PredP] when it precedes the (semelfactive) verb.

The starting point for Farkas and Kardos (2018) is the articulated VP structure, with an aspectual phrase (AspP) between vP and VP implicated in the aspectual interpretation of the predicate and determining a domain of aspectual interpretation (cf. also MacDonald 2008 or Travis 2010). The authors claim that POs, as non-subcategorized and non-thematic nominals, are base-generated in [Spec, AspP], with the verb undergoing V-Asp-v-T head movement.

Connecting the above restrictions with these syntactic accounts, I rely on Travis (2010), who claims that subcategorized and thematic internal arguments affecting the structure of the event of V are merged inside the VP but undergo movement to the specifier of AspP if they delimit the event of the verb. In addition, one may also argue that the same restriction in (17a) has to do with accusative POs banning accusative DP arguments but, instead, allowing non-accusative (e.g. elative) DP elements as in ivott egyet egy teából ‘(lit.) drank one.acc a tea.ela’. In addition, as verbal particles are claimed to be base-generated in a postverbal position, from which they move first to [Spec, AspP] and then to a vP-external position (Surányi ms), the co-occurrence restriction in (17b) receives syntactic support: POs and verbal particles compete for the same – base versus derived – syntactic position (for more on the claim that the verb-internal position is AspP and not PredP, as in É. Kiss 2008 or Surányi 2009, see Surányi ms). Moreover, according to Farkas and Kardos (2019a, 2019b), verbal particles and POs encode different (semantic) operators, and although they both derive a telic VP, they are associated with a different telic interpretation and they mutually exclude each other. Last, the restriction on stative verbs is a general property of ACOs cross-linguistically, which can be explained by the fact that COCs describe a wide range of (physical, mental or perceptual) actions involving temporal processes; therefore, they exclude states. The last two restrictions receive further semantic support: as POs introduce and modify a result (Section 6.2), they are incompatible with both delimited events (i.e., particle verbs) and states (cf. also Kiefer 2006 or Csirmaz 2008).

2.4 Measure phrases

Hungarian has a wide variety of accusative-marked temporal (18a) and spatial measure phrases (18b), measure phrases formed with the suffix -nyi (18c) as well as other measure phrases (18d):

They differ from the POs under consideration here not only because of the lack of their co-occurrence restrictions with another subcategorized (thematic) object (19a), a verbal particle (19b) or a stative verb (19c), but also because they can be definite and trigger the definite conjugation of the verb, as shown by the verb form futotta (versus the indefinite conjugation on futott) (20):

Such and similar sentences are perfectly acceptable because measure phrases express temporal/spatial duration, degree or extent. Also, since they do not impose any restriction on the argument structure of the predicate and do not have the properties of an argument, they do not merge as direct objects of the verb but are considered to be structurally case-marked adjuncts of the predicate. More precisely, due to restrictions on case-feature checking, temporal and spatial measure phrases are claimed to merge within vP, along with arguments and certain adjuncts. In case they appear preverbally, they move to a low preverbal functional projection above PredP (Csirmaz 2008). Hence, they can co-occur with verbal particles, situated in [Spec, PredP] (or even thematic objects) because they do not compete for the same syntactic position.

In sum, Hungarian POs under consideration in this paper differ both from thematic and cognate objects as well as accusative measure phrases.

With these preliminaries established, I turn to describing Hungarian as a strong satellite-framed language. This discussion is relevant as such a language is predicted to have ACOCs.

4.1 Previous proposals

Kiefer (1992, 1994, 2006 gives a general overview of the distribution, aspectual role and co-occurrence restrictions of the POs of class (a) and (b), and observes that they are not completely grammatical with the alatt ʻin’ time adverbial.

Piñón (2001), focusing mostly on egyet ‘one.acc’, gives a semantic analysis of Hungarian POs. As for its denotation, egyet ‘one.acc’ is argued to modify an event e of type R the runtime of which is a proper part of some time interval t, where t is contextually determined. Crucially, there is no other event e’ within t that is larger than or distinct from e. These two conditions ensure that event descriptions containing this PO are telic.

É. Kiss (2004) points out that the PO is a non-specific indefinite noun, which cannot be interpreted as the passive participant (Patient/Theme or Undergoer) of the event of the verb. She also emphasizes that what these constructions express is that the referent of the nominal in the syntactic subject position (i.e., the Agent) creates a realization or instantiation of the given event.

Csirmaz (2008) argues that these POs are situation delimiters. She points out that two telicizing elements – such as a PO and a telicizing verbal particle – generate a doubly delimited VP and hence violate Tenny’s (1994) Single Delimiting Constraint. This explains the ungrammaticality of *a fény fel-villant egyet ‘(lit.) the light prt-flashed one.acc’. Moreover, the presence of the PO excludes all other, non-telicizing verbal particles as well, as in *János el-mélázott egyet ‘(lit.) John prt-mused one.acc’, where the ungrammaticality is explained by the fact that particles require a specific object but the PO is non-specific.

Halm (2012) observes that there is no clear borderline between unaccusative and unergative verbs in Hungarian. More precisely, the fact that these verbs can appear with a verbal particle (e.g. a fény fel-villant ‘(lit.) the light prt-flashed’, behaviour typical of unaccusative verbs) but are also compatible with these POs (e.g. a fény villant egyet ‘(lit.) the light flashed one.acc’, behaviour typical of unergative verbs) cannot be considered as an argument for the adjunct status of POs and, consequently, for the unambiguous unaccusative behaviour of semelfactives but rather for their ambiguous unaccusative–unergative behaviour, which lends further support to Sorace’s (2000) unaccusative–unergative continuum.

Farkas and Kardos (2018) argue that the base-generated position of these POs is the specifier of a functional phrase in the verbal domain (AspP). Moreover, Farkas and Kardos (2019a, 2019b) claim that – in sharp contrast to verbal particles, result predicates and goal PPs – POs encode an aspectual operator that picks out a contextually specified, non-maximal subpart of the event in the denotation of the verbal predicate.

These prior works share two significant properties: they address only the POs of class (a) and (b), and they do not describe them as nominals fulfilling the function of the ACO in the language. Thus, an analysis of Hungarian POs as ACOs is still lacking.

4.2 A closer look at Csirmaz (2008)

Before articulating my proposal, I take a closer look at one of the previous studies, which addresses and rejects an account of these Hungarian POs in terms of CO ellipsis. Csirmaz (2008) assumes that the same result-based account can be invoked for English COs as well as for Hungarian egyet ‘one.acc’ and the POs of class (b), which introduce and modify a result. Later the author claims that, although some of their properties are similar, their distribution is different; therefore, an account of these POs in terms of CO ellipsis seems viable, but she argues against their CO analysis based on the following arguments: COs cannot always be elided in Hungarian, as shown in (24a); COs cannot always be added to a (modified) PO, as illustrated in (24b); and POs can also modify unaccusative semelfactives, as shown in (24c):

In addition, whenever a CO can alternate with a PO, it is assumed that the former is elided, as in (25):

Since this example involves ellipsis, it is not considered an instance of a PO construction.

With respect to the first argument, I believe that any analysis of Hungarian objects faces the analysis of the verbs hal ‘die’ and él ‘live’ and their COs as being exceptional in one sense or another.^[10] More precisely, these two verbs take a full (premodified) cognate nominal and not an accusative PO. Crucially, as an achievement verb associated with an inherent endpoint, hal ‘die’ cannot take any type of PO (see [26]) but either a verbal particle as in János meg-halt ‘(lit.) John prt-died’, where the particle overtly marks the inherent endpoint of the event, or a premodified CO as in János hősi halált halt ‘John died a heroic death’ (see also [24a]), where the CO cannot have the result interpretation and instead receives that of a manner adverb related to the modifying adjective (e.g. hősiesen ‘heroically’):

COCs built with the verbs die and live are among the very few cognate constructions that are exceptionally found cross-linguistically, even in languages that lack canonical ACOCs such as Hungarian. I maintain my position that Hungarian VPs built with these two verbs should be treated as somehow exceptional; therefore, I believe that the argument based on (24a), according to which COs cannot always be elided in Hungarian, does not provide a solid foundation for rejecting the CO (ellipsis) analysis of Hungarian POs more generally.^[11]

As for the second argument, I stress the idea that it is possible for COs to be added to an accusative adjective but only in the case of (transitive and) transitivizing COCs (cf. [27]). In such a case, the accusative adjective, denoting a subcategorized and thematic entity, is not an instance of a PO occurring in an ACOC and it does not share the syntactic properties of POs presented in Section 5 (e.g. as opposed to a PO, gyönyörűt ‘beautiful.acc’ can constitute the answer to a mit ʻwhat.acc’ question):

Such and similar instances involve ellipsis: in the absence of the CO rajzot ‘drawing.acc’, the accusative case marking appears on the adjective; see also Csirmaz (2008: 197).

When it comes to ACOCs, it is indeed true that, generally speaking, there is no free variation between the (modified) PO and the overt (modified) cognate nominal, and it is not the case that an overt cognate nominal is possible whenever a (modified) PO accompanies the prototypical unergative verb. In other words, I agree with Csirmaz (2008) that POs do not arise as a result of cognate nominal ellipsis as this type of approach is indeed untenable. But this does not mean that the cognate object analysis itself cannot be maintained. The approach I take here on POs is not in terms of ACO ellipsis but in terms of their being nominals that fulfil the function of the ACO in the language. The generalization that I make in this paper is that in Hungarian there are two COC patterns. On the one hand, there is the canonical – mostly transitive and transitivizing – COC pattern, also known from languages such as English, where the DP object is a semantic and morphological ‘copy’ of the verb. On the other hand, there is the non-canonical – mostly aspectual – COC pattern, not known from languages such as English, where the CO is expressed by egyet ‘one.acc’ or an accusative adjective. Whereas in the former case it is precisely the presence of the CO that defines the construction (which can exceptionally be elided as in [27] above), in the second case it is precisely the obligatory absence of the same type of object that defines the construction. It is this latter property that sets Hungarian apart from English and other languages that have canonical ACOCs, where the object is both semantically and morphologically related to the prototypical unergative verb it accompanies. In my opinion, a unified analysis of these two types of constructions does not and cannot lead to correct results and conclusions in Hungarian.

Three observations are related to the last argument, which is considered problematic because the single argument of unaccusative verbs behaves like the internal direct object argument of verbs used in a transitive configuration, hence it is difficult to account for the presence and syntactic position of the PO. From one perspective, these POs behave differently with (unaccusative or unergative) semelfactive and (unergative) activity verbs (see Piñón 2001). For instance, hányat ‘how.many.acc’ is acceptable with the former but not the latter verb type. In addition, egyet ‘one.acc’ can also precede the semelfactive verb and it can also be focalized or contrasted with another lexical element in a corrective focus position, implying its exclusion (e.g. an accusative numeral such as kettőt ‘two.acc’). All these facts can be explained by a possible lexical ambiguity of the PO, but I leave a deeper explanation for these facts to future work.

From a different perspective, the compatibility of semelfactive verbs with a PO sheds light on their ambiguous unaccusative–unergative behaviour (Halm 2012). Put differently, when a semelfactive takes a PO, it exhibits unergative behaviour, hence the PO should not be analysed as an adjunct but can occupy the vacant object position.

In addition, contrary to the traditional view (Hale and Keyser 1993; Levin 1993; Levin and Rappaport Hovav 1995), a cross-linguistic perspective on ACOs reveals that these objects can occur not only with unergative verbs but also with unaccusative, transitive or even ditransitive verbs in languages such as Modern Hebrew, Russian, Vietnamese or Arabic. The following unaccusative example from Modern Hebrew serves as illustration:

Hence, the co-occurrence of Hungarian POs with unaccusative semelfactive verbs should not be taken as an argument against the CO (ellipsis) approach per se.

In sum, although I agree with Csirmaz (2008) that the CO ellipsis analysis is untenable, I defend my position, through the strong counterarguments to her first and last claim, that the CO analysis itself – where POs are viewed as accusative elements that fulfil the function of the ACO – is justifiable. This is what the next section demonstrates.

5.1 The PO of class (a)

As the PO egyet ‘one.acc’ is not a subcategorized internal argument and does not denote a thematic entity that could be interpreted as being directly affected by the event of the verb (i.e., it is not the passive participant semantically labelled as Patient/Theme or Undergoer), it cannot appear in the derived subject position of a passive structure:

One of the first definitions of affectedness, given by Bolinger (1975), proposes the following constraint on passives with transitive verbs: the subject in a passive construction is considered to be a true patient, i.e., to be genuinely affected by the action of the verb, where affectedness is understood as a persistent change in the respective event participant (see also Beavers 2011). If the grammatical object in the active construction is not conceived as a true patient, there will be no corresponding passive structure. It is clear that in the above passive example egy ‘one’ cannot be construed as a true patient as it is not affected by the action of the verb.

Moreover, as noted by Doron (1983, 1986, among others, nominals appearing in non-referential positions are different from those that appear in referential and argument positions in that the latter can but the former cannot be quantified with strong determiners. As such, egyet ‘one.acc’ cannot occur with a strong determiner such as ez ‘this’, mindegyik ‘each’, a legtöbb ‘most’ or minden ‘every’, as illustrated below:

The incompatibility with strong determiners is one of the characteristic properties of predicative (that is, non-referential) nominals (Doron 1986).

As a non-thematic nominal, the PO is not assigned any theta-role by the verb; thus it does not have any semantic interpretation. Recall from Section 2.1 that Hungarian bare nominals, as predicative elements more generally, are considered to be non-referential. Hence, in the following example the essential difference between könyvet ‘book.acc’ and egyet ‘one.acc’ consists not in their being (non-)referential but in their being (non-)thematic. As such, the former is non-referential and thematic (Theme), and the latter is non-referential and non-thematic; therefore, they are not of the same thematic category:

In addition, the non-argument property of egyet ‘one.acc’ predicts that it cannot be pronominalized (Doron 1983). This prediction is borne out by the example below: although the demonstrative azt ‘that.acc’ has a [+/-specific] feature and can also refer back to a predicative constituent, even to predicates in copular sentences (cf. É. Kiss 2019), in the second sentence the demonstratives ezt ‘this.acc’ and azt ‘that.acc’ cannot stand for the PO egyet ‘one.acc’:

Interestingly, when the demonstratives refer back to a non-referential element, the verb is in the indefinite conjugation (cf. the verb form olvasott), the type of conjugation triggered by bare nouns and the indefinite PO egyet ‘one.acc’ (and other POs in general), as shown in (31) above. This is strongly connected to the indefiniteness restriction on ACOs more generally. However, when the same pronouns refer back to a referential argument, the verb is in the definite conjugation (cf. the verb form olvasta), the type of conjugation expected with definite objects in general as shown in (33):

Furthermore, egyet ‘one.acc’ cannot be modified by a restrictive relative clause (34a) and it cannot constitute the answer to a mit ʻwhat.acc’ question (34b).

With respect to its aspectual contribution, it has been noted that, similarly to English ACOs (see de Swart 2007; Horrocks and Stavrou 2006, 2010; Levin and Rappaport Hovav 1995; Macfarland 1994, 1995; Melloni and Masini 2017; Pereltsvaig 2002; Real Puigdollers 2008; Tenny 1994), egyet ‘one.acc’ effects a shift of aspectual character with respect to the corresponding unergative verb used on its own but it is not fully compatible with the alatt ‘in’ time adverbial; cf. also Kiefer (1992, 1994, 2006, Piñón (2001), Csirmaz (2008) or Farkas (2017):

It has been remarked in É. Kiss (2004) and Csirmaz (2008) that these constructions are not completely unacceptable with the alatt ‘in’ adverbial (see [36a]) and they are also compatible with the belül ‘within’ time adverbial (see [36b]). Moreover, if a (silent) Measure Phrase expressing a well-known distance is supplied contextually or is understood from the context (Dowty 1979; Zucchi 1998), the sentence becomes acceptable (see [36c]):

As the temporal adverbial test cannot be reliably used to diagnose the telicity of the verbal predicate with a PO, another standard telicity test is used to probe for (a)telicity in the case of this and other similar VPs. The conjunction test below shows that the sentence in (37) can only be interpreted to describe two eventualities, one in the morning and one in the afternoon. This kind of semantics is associated with telic event descriptions.

Conjoined atelic event descriptions, on the other hand, can also be interpreted as expressing a single eventuality, where the eventuality holds during both temporal intervals as one single eventuality, as shown below:

As the PO egyet ‘one.acc’ is not the accusative numeral egy ‘one’ but a nominal with a reduced form and meaning as in a grammaticalization process, it has no alternative it can be contrasted with. Therefore, it can be neither (contrastively) topicalized nor focalized or contrasted with another lexical element in a corrective focus position, implying its exclusion (e.g. an accusative numeral such as kettőt ‘two.acc’). In addition, as this PO does not have an adverbial counterpart, the discussion on its (possible) adverbial interpretation is irrelevant.

5.2 POs of class (b)

Similarly to egyet ‘one.acc’, members of this small and closed class of POs cannot appear in the derived subject position of a passive structure, as in (39):

In addition, as non-referential nominals, they cannot occur with a strong determiner (40):

As shown in the following sentence, both the bare noun pirítóst ‘toast.acc’ and the PO jót ‘good.acc’ occupy a preverbal position, they are both non-referential and trigger the indefinite conjugation of the verb (as per the indefiniteness restriction on ACOs more generally) as illustrated by the verb form reggelizett (versus the definite conjugation in reggelizte). However, in sharp contrast to the former, which is semantically interpreted as Theme, the latter is not interpreted as a theta-marked nominal:

Since the verb reggelizik ‘breakfast’, as a derived unergative verb, can also appear without an internal argument, the PO does not satisfy the argument structure properties of the verb, which would include getting a theta-role from it.

As predicate nominals in general cannot be pronominalized (Doron 1983, 1986), I predict that when these POs are replaced by a pronominal, they yield an unacceptable sentence. This prediction is borne out, as shown in (42), where the demonstratives should refer back to the PO:

Although POs of class (b) cannot be replaced by a (demonstrative) pronoun, in case of event anaphora – that is, anaphoric reference to events – the use of such and similar pronouns is, exceptionally, allowed (see also Mittwoch 1998; Real Puigdollers 2008 or Horrocks and Stavrou 2010). In such a case, the pronoun refers back neither to a referential and thematic nominal nor to the (pluralized) PO but to the entire (repeated) event expressed by the VP (on the event interpretation of the PO, see Section 6.1). In this respect, (43b) can follow (43a):

In such a case there is no contradiction between the singular pronoun ez ‘this’ in (43b) and the pluralized PO jókat ‘good.pl.acc’ in (43a). This is because the pronoun refers back not to a single (breakfast-eating) event but to the sum of repeatedly performed (breakfast-eating) events.

Similarly to egyet ‘one.acc’, POs of class (b) cannot be modified by a restrictive relative clause (44a), they cannot constitute the answer to a mit ʻwhat.acc’ type of question (44b) and they give rise to a VP which is slightly ungrammatical with the alatt ‘in’ adverbial (44c):

Although these few POs are treated as having some lexical content, they can be neither inserted in a sentence with contrastive topic (45a) nor focalized and contrasted with their direct antonym pair (45b):

Crucially, if the PO (egy) jót ‘(one) good.acc’ appears not with its direct antonym pair but one of the POs of class (c), for instance, the above sentences improve in felicity:

Next, if I look at the manner adverbial interpretation of these POs, I notice that they are not equivalent to (and hence cannot be replaced with) their adverbial counterpart. This is illustrated below, where the translations make it obvious that (47a) is not equivalent to (47b).

Although the adverbial counterparts of the adjectives hatalmas ‘huge’ (i.e., hatalmasan ‘in a huge way’) and óriási ‘gigantic’ (i.e., óriásin ‘in a gigantic manner’) sound odd, the adverbial counterpart of the last remaining PO of class (b) has the same behaviour as jól ‘well’ above: the PO-sentence in (48a) is not (necessarily) equivalent to the adverbial sentence in (48b):

That the PO jót ‘good.acc’ is not equivalent to the manner adverb jól ‘well’ is supported by the fact that the PO can co-occur with its adverbial counterpart within one and the same sentence:

As the PO jót ‘good.acc’ and the adverb jól ‘well’ fulfil different functions in the sentence, they do not display co-occurrence restrictions of any type.

This is further supported by the following sentence, where I want to test whether it is possible to insert a manner adverb with a meaning opposite to the meaning denoted by the adverbial counterpart of the adjective that the PO is built on (Real Puigdollers 2008). What I notice is that, as the PO jót ‘good.acc’ is not equivalent to the manner adverb jól ‘well’, the insertion of the manner adverb rosszul ‘badly’ does not lead to contradiction. Notice that in (50) the PO and the adverb appear in two distinct clauses and the second clause is introduced by the conjunction but (the sentence where the two appear in the same clause as in Mari rosszul táncolt egy jót ‘(lit.) Mary badly danced one good.acc’ may sound less natural in the language):

There is no contradiction in saying that Mary had a good dancing experience but she did not dance with talent.

Furthermore, the accusative PO jót ‘good.acc’ and the adverb jól ‘well’ can also be coordinated, as shown below:

At first sight, as coordination connects two (or more) conjuncts of the same syntactic category or semantic function, the grammaticality of this sentence might seem surprising. Although jót ‘good.acc’ and jól ‘well’ belong to two distinct word classes and differ not only in their morphological shape and meaning but also in distribution, the grammaticality of this sentence can be explained by assuming not the coordination of these two lexical items but the coordination of two clauses (É. Kiss, p.c.).

This section has shown that the syntactic properties of the POs of class (b) are similar to the behaviour of egyet ‘one.acc’ in the case of most syntactic tests but they also differ with respect to some other diagnostics.

5.3 POs of class (c)

Members of this large and open class of POs, as non-subcategorized, non-referential and non-thematic nominals, share most of the properties of the other POs. As such, they cannot appear in the derived subject position of passive structures:

In addition, they cannot co-occur with a strong determiner:

(54) shows that although pirítóst ‘toast.acc’ and gyorsat ‘quick.acc’, as bare singular nouns, share several properties (e.g. preverbal position, non-referentiality, indefinite conjugation triggered on the verb), they differ in at least one respect, namely in thematic properties, with the former being thematic and the latter, non-thematic.

As predicate nominals in general cannot be pronominalized (Doron 1983, 1986), I expect that when the POs of class (c) are replaced by a pronominal, they yield an unacceptable sentence. This prediction is borne out, as shown in (55), where the demonstratives are meant to refer back to the PO:

Although POs of class (c) cannot be replaced by a (demonstrative) pronoun, in case of event anaphora the use of such a pronoun is, exceptionally, allowed. Again, the pronoun in (56b) refers back to the entire (repeated) event expressed by the VP in (56a).

In addition, these POs cannot be modified by a restrictive relative clause (57a), they cannot constitute the answer to a mit ʻwhat.acc’ type of question (57b) and, with the in-time adverbial, they give rise to a sentence that is judged to be (slightly) ungrammatical (57c).

The tests of contrastive topicalization and focusing cast light on some differences between the PO egyet ‘one.acc’ and the POs of class (b), on the one hand, and the POs of class (c), on the other. The following two sentences are both felicitous:

As for the test of the adverbial interpretation, I remark that, in sharp contrast to the POs of class (b), which are not equivalent to their manner adverbial counterpart, the ones belonging to class (c) can be replaced with their adverbial correspondent.^[12] This is shown below, where the two sentences are semantically equivalent.

The semantic equivalence of the PO szelídet ‘gentle.acc’ and the adverb szelíden ‘gently’ is supported by the fact that they cannot co-occur within one and the same sentence:

As the PO and the adverb have more or less the same function (i.e., the PO contributes information about the manner in which the action denoted by the verb takes place), they mutually exclude each other.

This is further supported by the following pair of sentences, where I want to test whether it is possible to insert a manner adverb with a meaning opposite to the one denoted by the adverbial counterpart of the adjective that the PO is built on (Real Puigdollers 2008). What I notice is that, as the PO szelídet ‘gentle.acc’ is (or can be) equivalent to the adverb szelíden ‘gently’, the insertion of the adverb hevesen ‘heatedly’ leads to contradiction. Again, as shown below in (61), the PO and the adverb appear in two distinct clauses and the second clause is introduced by the conjunction but (the sentence where the two appear in the same clause as in A politikusok hevesen vitáztak egy szelídet ‘(lit.) The politicians heatedly debated one gentle.acc’ sounds grammatical in the language but it is as contradictory and pragmatically infelicitous as [61]):

To put it simply, there is a contradiction in saying that the politicians debated gently but in a heated manner.

Furthermore, the PO and its adverbial counterpart cannot be coordinated, as shown below:

This can be explained neither by the coordination of two lexical items nor by the coordination of two clauses. From this, I conclude once again that the PO and the adverb are subject to severe co-occurrence restrictions.^[13]

This section has shown that the syntactic properties of the POs of class (c) are similar to the behaviour of egyet ‘one.acc’ and that of the POs of class (b) in the case of most syntactic tests but they also differ with respect to some other tests.

5.4 Interim summary

Members of the three classes of POs share most of the properties of ACOs found in languages where they are semantically and morphologically related to the prototypical unergative verb they accompany. This is because ACOs and these POs are non-subcategorized, non-referential and non-thematic accusative nominals. The features common to the members of the three classes are reflected in the tests of passivization, quantification with a strong determiner, theta-role assignment, pronominalization, indefiniteness restriction, restrictive relative clause modification, mit ʻwhat.acc’ type of questioning and aspectual contribution. More importantly, the tests of contrastive topicalization, focusing and adverbial interpretation shed light on the most essential differences between them. All these syntactic features are summarized below:^[14]

As summarized in Table 1, the motivation behind the ternary division of Hungarian POs – instead of a binary one, which merges the POs of class (b) and (c) – is supported by syntactic evidence.

6.1 ACOs as events

According to Sailer (2010), the first empirical test for the event reading of a CO is the availability of a manner paraphrase. Relying on event variables to model the particular event reading of COCs (cf. Mittwoch 1998; Moltmann 1989; Parsons 1990), I illustrate the semantic representation of (63a), which is roughly equivalent to the corresponding intransitive sentence including the adverbial szelíden ‘gently’, in (63b):

This representation, which does not include tense or plural markings, expresses the proposition that there is an event e which is a debating-event; this event has the property of being gentle and the politicians are the participants of the event.

In addition, Mittwoch (1998) argues that (Hebrew) COs are neither thematic arguments of their predicates nor adjuncts but are realizations of the (non-thematic) Davidsonian event argument of the main predicate. I claim that (modified) POs in Hungarian are event arguments, which do not introduce an entity into the discourse and do not modify an already introduced entity but the event itself. I present my arguments with VPs built on unergative denominal verbs, with the object argument incorporated into V (Hale and Keyser 1993), as in fagyizik (<fagyi ‘icecream’) ‘eat an icecream’ or biciklizik (<bicikli ‘bicycle’) ‘ride a bicycle’. In such a case, the adjective on which the PO is built (e.g. jó ‘good’) modifies neither the referent of the nominal occupying the syntactic subject position nor the referent of the nominal in the syntactic object position (incorporated into the verb) but the event itself: the event of eating an icecream. This is sustained by the meaning of a sentence such as the following:

Proof of this comes from the fact that the scope of an inserted negative operator (nem ‘not’) can be the same adjective modifying either the subject or the incorporated object, without yielding semantic contradiction, but not the event itself. This is illustrated in (65):

This is further supported by the following non-pleonastic sentence, where the idea expressed by the linearly first clause does not entail the idea expressed by the linearly second clause:

The next evidence for the event interpretation of Hungarian POs comes from the reduplication of egyet ‘one.acc’. It is a well-known fact that in Hungarian verbal particle reduplication such as fel-fel ‘up-up’ in a sentence like János fel-fel dobta az érmét ‘(lit.) John up-up threw the coin.acc’ expresses the repetition of the event of the verb (cf. Lipták and Saab 2019). Similarly, egyet ‘one.acc’ can also be reduplicated as in egyet-egyet ‘one.acc-one.acc’ and in that case it signals the iteration of the event of the verb, i.e., that the event re-occurred an unspecified number of times during a certain time interval.

Without reduplication, the above verb refers to a single event, and with reduplication it refers to a series of (repeated) events.

POs of class (b) and (c) cannot be reduplicated but they can be pluralized. In such a case, the plural marking refers not to the plurality of an already introduced entity but to the plurality (or repetition) of the event itself. Proof of this comes from the fact that such a PO can also accompany a verb that takes a singular nominal as its syntactic subject and/or incorporated direct object (also lexicalized in the following example):

In such a case, PO pluralization has semantic scope over the denotation of the verb and the PO, and not just the PO itself. In other words, pluralization results in quantification over the event itself (i.e., event iteration as above) and not in quantification over the (quality of the) resulting state of the event (as in *‘The child participated in an extremely big/fantastic bike-riding event’).

6.2 ACOs as results

In addition to their eventive interpretation, Hungarian POs can also denote resultant objects whose referents are ‘produced’ by the actions expressed by the verb. In this case, the verb describes first what kind of action is performed by the subject referent, and then the PO describes what was produced by this action or, more precisely, describes the ‘product’ resulting from the performed action. As such, the PO accompanying an emission verb has a polysemous interpretation: it not only refers to the event itself but it also identifies its result (i.e., a concrete manifestation); hence it can denote an emitted sound or a (bodily) substance. Also, as an effected entity, it manifests the interesting property of being co-extensive with the event of the verb: such nominals refer to entities which come into existence through the unfolding of the event but they cease to exist with the end of the same event (cf. Horita 1996).

First, with emission verbs – where the term ‘emission’ is taken in a larger sense and thus may refer to ‘emission’ of facial expression as in mosolyog ‘smile’, emission of sound as in kiált ‘shout’ or emission of sound and bodily substance as in tüsszent ‘sneeze’ – the PO denotes, first and foremost, an effected entity as evidenced by the interpretation of the following sentence:

(69) can roughly be paraphrased by a sentence with the morphologically simple nominal mosoly ‘smile’, which refers to a particular constellation of the facial muscles (ʻMary smiled, which resulted in a wide smile’) and not by a sentence with the morphologically complex, event-denoting -ás noun mosolygás ‘smiling’, which refers to the action of smiling (ʻ≠Mary smiled, which resulted in a wide smiling event’).

Second, as POs have an ‘introduction/generation-of-an-event’ interpretation (É. Kiss 2004), they are incompatible with creation verbs, which have a ‘generation-of-a-referential-entity’ interpretation (cf. also Marantz 2005). To put it in more formal terms, ACOs are often characterized as objects of result or effected objects, that is, they do not denote pre-existing ‘entities’ affected by the action denoted by the verb; rather, they are ‘entities’ created by the action of the verb. That is, the object in this case is interpreted as if it were linked to some event, similar to an event of creation. This is exactly why they are not compatible with verbs of creation, which in their basic sense involve an effected entity, which is syntactically analysed as a subcategorized internal argument. This explains the ungrammaticality of (70):

However, if the creation meaning is backgrounded and the event interpretation is foregrounded by the insertion of some suffixes, the verb becomes compatible with the PO:

In sum, Hungarian POs can also denote resultant objects whose referents are ‘produced’ by the actions expressed by the verb. I have demonstrated this with emission and creation verbs.

This section has presented clear evidence in favour of the idea that Hungarian POs can refer either to the event (and have an eventive interpretation) or to the entity that results from the verbal event (and have an effected object interpretation). In sum, both event and effected object readings are attested in the PO constructions of this language.