From relative proadverb to declarative complementizer: the evolution of the Hungarian hogy ʻthatʼ

1 Introduction

Declarative complementizers of several languages, among them the Italian che (Lehmann 1984: 393), the French que (Brachet and Kitchin 2015), the Russian čto (Meyer 2017), the Chalcatongo Mixtec xa=, the Thai thîi, or the Early Biblical Hebrew she/asher (Heine and Kuteva 2002: 254) are claimed to have developed from relative pronouns. Recently it has been argued about Germanic that-type complementizers, among them the German dass and the English that, that they derive from relative pronouns (instead of demonstratives, as traditionally assumed), and their source structure – similarly to that of the Hittite complementizer kuit and the Old Indian complementizer yád (Lühr 2008) – was the correlative construction (Axel-Tober 2017). The correlative construction, widespread in Proto-Indo-European (Auderset 2020; Luján 2009), consists of a headless relative clause, and a main clause containing a demonstrative or a definite NP that is related to the relative clause and the relative pronoun anaphorically, as shown by the Hungarian example in (1).

The initial phase of the hypothesized evolutionary process, however, is unclear. The construction to which the Germanic that-type complementizers can be traced back is not a canonical correlative structure any more; it is a so-called explicative construction of the type Mary knows that, that Peter is lying. The explicative clause relates to the main clause like a correlative sentence in as much as it is adjoined to the main clause, and it specifies the content of the demonstrative in the main clause. Its relative clause status, however, is doubtful. The pronoun introducing it does not bind a variable; it is a head in complementizer position. Haegeman (2012) and Axel-Tober (2017) assume that the clause contains an empty operator in Spec,CP binding the event variable; however, there is no clear anaphoric relation between the main clause demonstrative and this variable.

The present paper supports the proposed derivation of Indo-European that-type complementizers by presenting a similar developmental path whose early phases can be reconstructed more completely, that of hogy ‘that’, the Hungarian general complementizer cognate with the relative pronoun hogy ‘how’. It traces hogy back to a canonical correlative construction, documenting the subsequent stages of its evolution from a relative operator binding a variable in a correlative adjunct clause to a general complementizer subordinating a clausal argument to a matrix predicate. The analysis of the evolution of hogy suggests that the explicative construction identified as the source structure of that-type complementizers may actually represent an intermediate stage in their evolution, which started in a canonical correlative construction.

The paper is organized as follows: Section 2 argues that in Proto-Hungarian and Proto-Ugric, languages with non-finite subordination, the first complex sentence type involving finite subordination was the correlative construction. Section 3 claims that hogy grammaticalized into a general complementizer in the clausal complements of verbs of communication, and follows their evolution from paratactic complement clauses announced in the quoting sentence by a cataphoric úgy ‘so’ (Section 3.1), via adjunct clauses linked to the quoting sentence by hogy, the relative counterpart of úgy, in quasi-correlative (explicative) constructions (Section 3.2), to complement clauses integrated into the argument structure of the quoting verb (Section 3.3). Section 3.4 discusses the emergence of a demonstrative pronominal associate of the argument clause in the matrix sentence. Section 4 surveys the spreading of hogy from the complement clauses of verbs of communication to other contexts. Section 5 is a summary.

2 The emergence of finite subordination in the Ugric languages: the correlative construction

In order to reconstruct the emergence of the first Hungarian complex sentence type involving two finite clauses linked in an asymmetric relation, we have to go back beyond the documented history of Hungarian beginning in 1192; we also have to study subordination in Khanty and Mansi (also known as Ostyak and Vogul), the conservative sister languages of Hungarian in the Ugric branch of the Uralic family. Hungarian was subject to heavy Indo-European influence after 895, when the Hungarian tribes settled in the Carpathian Basin. They presumably absorbed the Slavic population of the area, which lead to a directionality shift from the Uralic SOV word order to SVO. Khanty and Mansi avoided strong Slavic influence until the 2nd half of the 20th century; they only recently have started showing signs of the loosening of head-finality. The properties of older Khanty and Mansi texts that are also present in Old Hungarian in gradually decreasing proportions are likely to represent Proto-Ugric heritage which was also shared by Proto-Hungarian when it split off the Ugric branch of Uralic.

Khanty and Mansi, like many strictly SOV languages, have had no finite subordination until recently. Argument and adjunct clauses are mostly expressed by a great variety of non-finite constructions, which may have an independent subject of their own eliciting agreement on the non-finite verb (Csepregi 1998/2011; Csepregi and Gugán 2020; Filchenko 2007; Gulya 1966; Kálmán 1976; Nikolaeva 1999; Riese 2001). The propositional complements of quoting predicates and predicates reporting propositional attitudes, which can have tense and mood independent of those of the reporting predicate, tend to be expressed by independent sentences in paratactic constructions. Example (2) illustrates both possibilities. The reporting S1 and the reported S2 and S3 are independent sentences, whereas S4, the subject clause of S3 is a participial expression.

Relative clauses are prenominal gap-relatives displaying no relative pronouns:

The only type of complex sentence involving two asymmetrically linked finite clauses in older Khanty and Mansi texts is the correlative construction (see Dékány et al. 2020, and the annotated documents of the Ob-Ugric Database). The correlative construction, found mostly in OV languages (Keenan 1985), consists of a free, headless relative and a main clause (Bhatt 2003; Dayal 1996; Grosu and Landman 1998; Lipták 2009, 2012).^[2] The relative clause occupies a peripheral position; in the canonical case it precedes the main clause. Lipták (2012) argues based on reconstruction and locality effects that it is base-generated adjoined to the main clause. The correlate of the relative clause, i.e., the element in the main clause associated with it, is a demonstrative or a personal pronoun. Either the relative pronoun, or the demonstrative, or both can have their restrictor spelled out, as illustrated by the Hungarian sentences in (4a–c). The examples also illustrate the lack of reconstruction effects: the referring expression contained in the correlative clause is freely coindexed with the subject of the main clause.

Since the relative clause does not modify a nominal, it can contain more than one relative pronoun:

The relative pronoun expresses maximality; it corresponds to a free choice expression; thus amelyik diák in (4) means ‘any student’, and who in (5) means ‘anybody’.

The correlative constructions of Khanty and Mansi, in essence, share these properties. They display the canonical ‘relative clause, main clause’ order. The definite pronoun or proadverb in the main clause can be pro-dropped or implicit. The pronouns in the relative clauses are actually underspecified indeterminate pronouns rather than relative pronouns – as the traditional, pre-Russification versions of these languages only had prenominal relative clauses with a gap instead of a relative pronoun. Indeterminate pronouns provide a free variable that is bound by an overt or implicit existential, interrogative, etc. operator. Correlative clauses have been claimed to contain an implicit conditional or generic operator (Belyaev and Haug 2020; Bende-Farkas 2014; Keshet 2013), which, too, could serve as the binder of the indeterminate pronoun.

In traditional Khanty and Mansi texts, the correlative constructions represent the first step in the emergence of finite subordination.^[4] In the case of Hungarian, we cannot document this stage; the first surviving documents already show flexible SVO word order, and an increasing proportion of finite subordinate clauses introduced by the complementizer hogy. Nevertheless, the abundance of non-finite subordination and parataxis, and the abundance of correlative constructions in early Old Hungarian documents (decreasing gradually throughout the Old Hungarian period) suggest that the language barely left behind the stage represented by older Khanty and Mansi texts. The pronouns of the relative clauses of correlative constructions must still be indeterminate pronouns bound by an implicit operator rather than relatives, as they are still often supplied with the existential vala- ‘some’ prefix (Bende-Farkas 2014):

Whereas Khanty and Mansi correlative constructions always begin with the relative clause, the restructuring of Hungarian grammar from head-final to head-initial gradually led to the preponderance of the reverse ‘main clause – relative clause’ order. In ‘main clause – relative clause’ constructions, the pronoun of the relative clause almost never bears the existential vala- prefix, which suggests that it is not an indeterminate pronoun any more but a true relative pronoun in Spec,CP.

In (10b), the pro-drop of the demonstrative is licensed by object–verb agreement.

The question arises whether these ‘main clause – relative clause’ complexes can still be regarded as correlative constructions, given that – at least in Modern Hungarian – the relative clause in this construction cannot contain more than one relative pronoun. The relative clause is clearly adjoined to the main clause in examples like (10a–c); what is unclear is whether this is its base position, or it has been extraposed from the demonstrative in the main clause. I hypothesize that initially the construction was a reverse correlative construction; the reanalysis of the relative clause as the extraposed modifier of the main clause pronoun or proadverb was a subsequent step.

In Old Hungarian, we find correlative clauses associated with all types of arguments and adjuncts. In manner correlative constructions, the relative proadverbs hogy ‘how’ and mi-ként, the essive-marked demonstrative meaning ‘like what; in what way’ co-occurred with the demonstratives úgy, így ‘so’:

The úgy … hogy … ‘so … that …’ pair is also common in complex sentences involving a finite consecutive clause of manner or degree:

Owing to the frequency of the correlative construction and of the consecutive construction illustrated in (12), the demonstrative – relative pronoun pairs listed below must have been set correlates in Old Hungarian.

As will be argued below, the association of hogy with így/úgy played a crucial role it the development of hogy into a general complementizer.

3 The integration of reported propositions

3.1 From independent sentence to (correlative) adjunct

The evolution of the subordinating function of the relative pronoun hogy must have started in the propositional complements of verbs of communication, and it must have been due to the proadverb így/úgy accompanying these verbs.

Recall that verbs of communication, taking propositional complements with tense and mood independent of those of the main predicate (Lohninger and Wurmbrand 2020), occur in paratactic constructions in Khanty and Mansi, apart from occasional cases of non-finite subordination. Direct quotations are still typical in Old Hungarian, as well. The quoting predicate is often accompanied by the cataphoric proadverb úgy ‘so’ or its proximal equivalent így. As argued by Munro (1982) and discussed by É. Kiss and Gugán (2021), ‘say’ verbs are only weakly transitive; they display various intransitive characteristics crosslinguistically. The quoted material is often treated as oblique; e.g., in some languages it can be questioned by a manner wh-adverb, as illustrated by the Italian and Khanty examples in (14).

(14)

a.

Come	avete	detto?	(Italian)
how	have-pst.2sg	said
‘What have you said?’
(Munro 1983: 315)

b.

jǝγ-ǝm	qŏłnǝ	nüŋ-ati	jast-ǝł?	(Khanty)
father-1sg	how	you-dat	say-prs.3sg
‘What does my father tell you?’
(Paasonen 2001, OUDB 1315)

English examples of type (15), containing a verb of communication and a complement clause introduced by how, discussed by Legate (2010) and Van Gelderen (2015), may be manifestations of the same phenomenon:^[7]

(15)

They told me how the tooth fairy doesn’t really exist. (Legate 2010: 121)

Such examples, not accepted by all speakers in English, have been shown to occur also in Breton (Willis 2007), French, Spanish, Greek, Hebrew, Mandarin Chinese, and Warlpiri (Legate 2010).

In Hungarian, the grammaticalization of úgy mond ‘so says’ into the discourse particle úgymond ‘so to speak’ (Dömötör 2008) is indicative of a former (marginally still existing) selectional relation between mond and úgy. As the Old Hungarian examples below indicate, úgy was selected not only by mond but by all kinds of verbs of communication. Notice that the personal pronouns of the reported propositions display no indexical shift, they preserve the perspective of the subject of the verb of communication, which is evidence of the independence of the reporting and reported sentences.

(16)

a.

istèn	ug mond-ot	Tiztel-l-èd	te    atʼ-ad-at	&
god	so say-perf.indef.3sg	respect-imp-obj.2sg	you father-2sg-acc	and
te	ań-a-d-at
you	mother-2sg-acc
‘God said so: Respect your father and your mother!’
(München C. 1416: 21v)

b.

Es	te	vgy	felel-y :	Adkoztatt-ak	kewzybe	melto	vagy
and	you	so	answer-imp.2sg	cursed-pl	among	worthy	be.2sg

zamlal-tatt-n-od

count-pass-inf-2sg

‘And answer so: You are worthy of being counted among the cursed.’

(Jókai C. 1370: 33)

c.

vgy	paranczolt	my-nekw̋nk	wr-vnk:	Vetee-lek	teghed	nepek-nek
so	ordered	we-dat.1pl	lord-1pl	threw-1sg	you.acc	peoples-dat

vylagoss-a-ra

light-poss-all

‘Our lord ordered us so: I have thrown you to the light of peoples.’

(Jordánszky C. 1516: 755)

d.

ug scola-noc	wylag-noc	kezdet-u-i-tul	fugua	roht-onc	ez
so spoke-3pl	world-dat	beginning-poss-pl-abl	since	upon-1pl	this

nem	lev-t	wal-a	hug	scuz	lean	fio-t	sciul-hes-s-en
not	occur-perf.3sg	be-pst	that	virgin	girl	boy-acc	bear-possib-sbjv-3sg
‘They spoke so: It has not occurred to us since the beginning of the world that a virgin girl can bear a son.’
(Königsberg Fragment middle of 14th c.)

e.

A	meǵ_èmleitet	ꝓpheta	ig	bèzellet	Istèn	haragos	and	go̗zo̗dèlmès
the	mentioned	prophet	so	talked	God	angry	and	victorious
‘The mentioned prophet talked so: God is angry and victorious.’
(Vienna C. 1416/1450: 2/257)

In Hungarian, the emergence and spreading of finite complement clauses proceeded parallel with the restructuring of Hungarian grammar from head-final to head-initial (Bacskai-Atkari and Dékány 2014; É. Kiss 2013). Whether the emergence of finite subordination was a consequence of the VO order (as predicted e.g. by Hawkins [2014]), or it was a direct consequence of Indo-European influence, finite subordination was spreading continually throughout the Old Hungarian period – as is shown e.g. by the growing proportion of finite subordinate clauses in subsequent translations of the four Gospels (É. Kiss 2013, 2014: 30–31). When the pressure to integrate propositional complements as finite subordinate clauses reached communicative verbs, the only means that Hungarian had to link two finite clauses in a hypotactic structure was the correlative construction, with a relative pronoun introducing the complement clause. Úgy, the demonstrative proadverb accompanying quoting predicates, called forth the relative pronoun hogy ‘how’ as its correlative pair, as illustrated in (17):

(17)

a.

vǵ	mond	zenth	Gergel	doctor,	hoǵ	az	ǫrdǫg	ez	fǫld-et
so	says	Saint	Gregory	doctor	as	the	devil	this	earth-acc
kereng-i
circle-obj.3sg
‘So says doctor St Gregory as/that the devil is circling this earth.’
(Bod C. 1500–1525: 9r)

b.

Ky-k-nek	en	akkoron	vgy	feleleek,	hogy	nem	sokas-ok	az
who-pl-dat	I	then	so	answered	as	not	habit-poss	the

romayak-nak	vala_meely	ember-t	halar-ra	ad-ny.’
Romans-dat	some	man-acc	death-to	give-inf
‘Whom I so answered then as/that the Romans are not in the habit of handing people over to death.’
(Jordánszky C. 1516: 789)

c.

így	szóllyatok	nék-i:	hogy	az    Úr	kíván-nya	en-nek
so	call-imp.2pl	dat-3sg	as	the Lord	wish-obj.3sg	this-dat
szolgálatt-y-át
service-poss-acc
‘Appeal to him so as/that the Lord wishes its service.’
(Káldi 1626: 976)

The documents containing these sentences are from later times on an average than the sources of the paratactic constructions in (16). Nevertheless, it is important to emphasize that every stage in the process leading to the grammaticalization of hogy into a general complementizer already emerged in the Proto-Hungarian period. Hogy as a general complementizer had been “ready” by the time of the first surviving Hungarian documents, but the newly evolving stages in the grammaticalization path did not overwrite the previous stages. Every stage has survived, with the constructions representing more recent stages spreading slowly, and those representing the earlier stages gradually losing ground.

The sentences in (17) are only formally correlative constructions. They are explicative structures in the sense that the subordinate clause explicates the adverb úgy ‘so’ of the main clause; it explicates what the agent of the main clause communicated; how the communication sounded. Úgy plays the same role in these constructions as the main clause demonstrative in constructions of the type Mary knows that, that Peter is lying in the Germanic languages, where the complement of weakly transitive communicative verbs is construed as an object. The pronoun hogy is not a relativizer; it merely functions as a linking element, a subordinator. It adjoins the clause containing the reported proposition to the main clause, whereby it establishes the same structural relation between the two clauses that is attested in a correlative construction – see (18), the tree diagram of example (17a). Crucially, the matrix verb bears no object agreement morpheme, which indicates that embedding is shallow; the subordinate clause has not yet evolved into an object clause.

(18)

In these constructions hogy is not a wh manner adverb moved to Spec,CP any more; it is already a complementizer base-generated in C – as argued before by Bacskai-Atkari and Dékány (2014) and by Bacskai-Atkari (2016).^[8] Hogy as a wh-adverb in Spec,CP is a phrase that can be replaced by the PP mi-ként ‘like what’, a wh-pronoun supplied with the essive case suffix (see 11a), but hogy as a complementizer has no phrasal alternative. It also lacks the lexical manner feature of hogy in Spec,CP. The grammaticalization from a phrase in Spec,CP to a head in C has been claimed to be a typical step in the evolution of complementizers cross-linguistically (Roberts and Roussou 2003: Ch. 3), a manifestation of the relative cycle (van Gelderen 2011), and has also been pointed out in the case of the Hungarian complementizers ha ‘if’, mert ‘because’ and mint ‘as’ (Bacskai-Atkari 2016; Bacskai-Atkari and Dékány 2014). The evolution of the German complementizer dass may also have followed a similar path; its position in C in Old and Middle High German explicative constructions is claimed to be the result of its grammaticalization from a relative wh-phrase with phi-features, landing in Spec,CP, to a head without phi-features base-generated in C (Axel-Tober 2017: e42).

Úgy had a crucial role in evoking hogy as a clause linker, but once hogy started functioning as a complementizer in C, its presence in constructions of type (18) was not essential any more. (19a,b) are complex sentences containing an adjunct clause introduced by hogy, where the main clause contains no cataphoric úgy. In (19b) hogy introduces a wh-question, which is further evidence against its relative pronoun status and for its complementizer role. The interrogative wh-phrase occupies Spec,FocP, as is the rule in Hungarian syntax.

(19)

a.

my	vymadsag-on-bam	ker-y-enk	my	vronk	tol	íesus
we	prayer-1sg-in	ask-sbjv-1pl	our	lord	from	Jesus

cristus-tol	hogy	meltol-y-on	nek-enk	mutatt-ny-a …	nek-y
Christus-from	that	condescend-sbjv-3sg	dat-1pl	show-inf-3sg	dat-3sg

kelew	ut-at
needed	way-acc
‘In my prayer, we shall ask our Lord Jesus Christ that he shall condescend to show us the way needed by him.’
(Jókai C. 1370: 6)

b.

az	vt-ban	veto̗ko̗d-t-ec val-a	hog	ki	o̗	ko̗zo̗tto̗c	nagob
the	way-in	compete-perf-3pl be-pst	that	who	they	among	greater
vol-a
be-pst.3sg
‘They had competed on the way [that] who was greater among them.’
(München C. 1416/1466: 45r)

3.2 From adjunct to argument clause

The emergence of agreement between verbs of communication and their complement clauses represents a further step in the evolution of finite subordination. Though the very first Hungarian document from 1192 already contains an embedded question eliciting object–verb agreement on the matrix verb (20), and there are sporadic examples from the 14th and 15th centuries, as well, agreement between verbs of communication and their propositional complements only became general in the 16th century (21a–b).

(20)

Ge	mund-oa	nek-í	[meret	nū	en-e-ẏc]
but	say-pst.obj.3sg	dat-3sg	why	not	eat-cond-3sg
‘But he told him why he shall not eat.’
(Funeral Sermon 1192)

In complex sentences containing an object clause eliciting verbal agreement, the demonstrative úgy is mostly absent – as is the case in (21a–c).

(21)

a.

meg	ielent-e	az	ersek	az	baratok-nak	hog	az
prt	report-pst.obj.3sg	the	archbishop	the	friars-dat	that	the

ierusalem	bely	zerzetessek-nek	vol-na	zokas-ok	ez	keppen
Jerusalem	from	monks-dat	be-cond.3sg	habit-3pl	this	way

tyztel-ny …	zvz	maria-t [9]
honor-inf	Virgin	Mary-acc
‘And the archbishop reported to the friars that the monks in Jerusalem were in the habit of honoring the Virgin Mary this way’
(Horvát C. 1522: 119)

b.

Parantsoll-y-ad	hoǵ	az	én	kęt	fiai-m	uͤll-ye-nec	a	te
command-imp-obj.2sg	that	the	I	two	sons-1sg	sit-sbjv-3pl	the	you
ország-od-bā
country-2sg-in
‘Command that my two sons live in your country’
(Károli 1590: 20r)

When úgy or its interrogative counterpart miképpen ‘how’ occurs alongside a verb of communication bearing an object agreement morpheme cross-referencing the complement clause, it functions as a manner adjunct:

(22)

a.

Mert	vgy	rendel-t-e	val-a,	hogy	oͤ	foͤld-en	oda	akar-na
because	so	order-perf-obj.3sg	be-pst	that	he	land-on	there	want-cond.3sg
men-ni.
go-inf
‘Because so did he order that he wanted to go there on land.’
(Heltai 1561: O1v)

b.

Miképpen	mond-gyác	az	irástudóc	hogy	a’	Christus
how	say-obj.3pl	the	scripturists	that	the	Chirst
Dauid-nac fi-a?
David-dat son-poss
‘In what way do the scripturists say that Christ is David’s son?’
(Heltai 1561: M6v)

Complement clauses eliciting object agreement on the matrix verb bind a trace in the matrix VP, as represented below in the tree diagram of example (22b):

(23)

Whereas mond has become obligatorily transitive, some verbs, e.g. telefonál ‘telephone’, ír ‘write’, still can take their propositional complement either with agreement, as an object clause, or without agreement, as an adjunct clause. As expected, extraction (long focusing or wh-movement) is only allowed from object clauses eliciting agreement on the matrix verb:

(24)

a.

Ír-t	/Ír-t-a	János,	hogy	össze-vesz-ett	a
write-pst.3sg	/write-pst-obj.3sg	John	that	out-fell-pst.3sg	the
feleség-é-vel.[10]
wife-3sg-with
‘John wrote that he fell out with his wife.’

b.

Ki-vel i	*ír-t	/ír-t-a	János,	hogy	össze-vesz-ett ti?
who-with	write-pst.3sg	/write-pst-obj.3sg	John	that	out-fell-pst.3sg
‘Who did John write that he fell out with?’

3.3 From argument clause to clausal associate of a demonstrative pronoun

The next step in the evolution of finite subordination is the emergence of a main clause demonstrative coindexed with the complement clause. Initially, both the proximal and the distal versions of the demonstrative occurred, but proximals later disappeared.

(25)

a.

az-t	mon-t-a	isten	[ hog	gÿarapog-gÿa-tok	es	sokassul-ÿa-tok]
that-acc	say-perf-obj.3sg	god	that	increase-imp-2pl	and	multiply-imp-2pl
‘God said that that you shall increase and multiply.’
(Sándor C. 1500–1525: 17r)

b.

De akar-om	hoǵ	tug-ǵā-tok	ez-t	[ hoǵ	minden	firfiu-nak
but want-obj.1sg	that	know-sbjv-obj.2pl	this-acc	that	every	man-dat

fei-e	az	Chriſtus]
head-poss	the	Christ
‘But I want that you should know this that every man’s head is Christ.’
(Sylvester 1541: II.33v)

In propositions integrated as object clauses eliciting verbal agreement, deictic elements always undergo indexical shift – as is illustrated by (26), where the 1st person singular subject of the reported sentence (I am God’s son) became 3rd person. Notice, however, that Hungarian has no “sequence of tenses”; the tense of the embedded proposition is not relativized to the matrix tense.

(26)

Mert	az-t	mond-ā	hoǵ	fi-a	uͤ	az	isten-nek.
because	that-acc	say-pst.obj.3sg	that	son-poss	he	the	god-dat
‘Because he said that he is the son of God.’
(Sylvester 1541: I. 46r)

The emergence of pronominal associates of clausal complements must have been motivated by the fact that the complement clause often needed to be represented in the left-peripheral topic or focus position of the main clause – see (27). Whereas the topicalization of a complete clause is merely cumbersome, the focus slot is completely inaccessible to heavy constituents.

(27)

ćzak [FocP	AZ i	legÿen k	[VP vÿgasag-tok tk ti]]	[CP	hoǵh	az	tẃ
only	that	be.imp.3sg	joy-2pl		that	the	you

neu-etek	ffel	ÿrattat-ok	meń	orzag-ban]i
name-2pl	up	written-pl	heavenly	country-in
‘Your joy be only that: that your names are inscribed in heaven.’
(Könyvecse 1521: 2r)

The exact nature of the relation between the subordinate clause and the demonstrative coindexed with it in examples like (26–27), represented in (28), is a debated issue.

(28)

A clue to the derivation of structure (28) is provided by the fact that the demonstrative turns the complement clause into an extraction island. Compare:

(29)

a.

( Az-t)	mond-ják,	[ hogy	Mari	férjhez	megy	János-hoz]
that-acc	say-obj.3pl	that	Mary	married	gets	John-to
‘They say that Mary will get married to John.’

b.

* Az-t	ki-hez i	mond-ják,	[ hogy	Mari	férjhez	megy ti]?
that-acc	who-to	say-obj.3pl	that	Mary	married	gets
Intended: ‘Who is Mary said to get married to?’

cf. c.

Kihez i mondják,	[ hogy Mari férjhez megy ti]?
‘Who is Mary said to get married to?’

The ungrammaticality of (29b) follows whether the embedded clause is the extraposed complement of the pronoun, or an adjunct coindexed with it. In the former case, the ungrammaticality of (29b) follows from the Complex NP Constraint; in the latter case, it follows from the adjunct island constraint (more precisely, from the Condition on Extraction Domain).

If úgy accompanying an object clause triggering verbal agreement is a mere manner adjunct that does not affect the argument status of the embedded clause, then it is not expected to block wh- or focus extraction. In fact, it does (30a) – but only because úgy itself needs to be focused whereby it occupies the potential landing site of the extracted element. If the long movement targets the focus slot of a higher clause, úgy does not prevent it (30b).

(30)

a.

(*Úgy)	János-hoz i	hall-ott-ák	(*úgy),	hogy	Mari	férjhez	megy ti.
(so)	John-to	hear-pst-obj.3pl	(so)	that	Mary	married	gets
‘It is to John that they heard that Mary would get married.’

b.

János-hoz i	mond-ják,	hogy	úgy	hall-ott-ák,	hogy	Mari
John-to	say-obj.3pl	that-acc	so	hear-pst-obj.3pl	that	Mary
férjhez	megy ti.
married	gets
‘It is to John that they say that they heard that Mary would get married.’

4 The spreading of finite subordination with hogy

It has been argued above that hogy assumed the function of a general complementizer in the context of verbs of communications. Before Hungarian developed the grammatical means of finite subordination, these verbs took their propositional complements in paratactic constructions where the verb of communication was often accompanied by a cataphoric úgy ‘so’. When the reported sentence came to be integrated into the reporting clause, úgy evoked its correlative pair, hogy, as a linking element. Actually verbs of propositional attitudes such as tud ‘know’, hisz ‘believe’, and verbs of perception such as lát ‘see’ and hall ‘hear’ also provided similar contexts. The reason why their contribution to the evolution of hogy may have been less significant is that they were used much less frequently than reporting verbs. For example, the number of the occurrences of the inflected forms of mond ‘say’ in the paleographically normalized part of the Old Hungarian Corpus is 767; this number is only 69 in the case of tud ‘know’, 85 in the case of lát ‘see’, and 12 in the case of hall ‘hear’.

Whereas verbs of communication must have had a distinguished role in the evolution of hogy, once hogy was established as a complementizer, its use spread quickly to other contexts as well, among them the complements of attitude predicates (31a) and predicates of perception (31b):

Hogy also appeared with active and passive verbs taking a propositional subject:

The emergence of a demonstrative coindexed with the subordinate clause also enabled hogy clauses to function as complements of verbs selecting an argument with a lexical case, and as complements of postpositions. In (33), the demonstrative coindexed with the complement clause serves as the lexical base of the ablative case suffix selected by the matrix verb:

In (34), the demonstrative functions as the PP-internal complement of a postposition. Postpositions are unable to combine with a CP directly; they subcategorize for a nominal projection – presumably because Hungarian PPs are partially grammaticalized possessive constructions. In Old Hungarian, their grammaticalization was less complete than it is today; the demonstrative in (34) bears the dative case of possessors, and the possessum (the postposition) bears possessive agreement. (In Modern Hungarian, dative marking on the complement of P and possessor agreement on P only appear if the complement of P is extracted; in the default case, both are unmarked.) The clausal associate of the demonstrative is adjoined to the topicalized PP.

Certain types of adjunct clauses, e.g., clauses of purpose, are also introduced by hogy (35a,b). The optional pronominal associate of the purpose clause bears the causal-final suffix -ért ‘for’ (35b).

In 16th century documents, hogy occasionally even shows up as a general complementizer in relative clauses, preceding the relative operator (Dömötör 1992: 674; Galambos 1907). Bacskai-Atkari and Dékány (2014: 292–222) and Bacskai-Atkari (2016) argue that the relative clause contains two CP layers, with hogy in the higher C position and the relative operator in the specifier of the lower CP.

5 Summary

The Hungarian complementizer hogy is likely to have grammaticalized in the context of predicates of communication. The analysis of Old Hungarian data (from 1192 onwards), supplemented by relevant data of the conservative sister languages of Hungarian, has revealed five stages in the evolution of hogy. Although manifestations of the first four stages have been present in the language since the first written documents up to now, the decreasing proportions of the earlier versions, and the documented late appearance of stage 5, as well as language-external parallels (such as the primacy of the correlative construction in the evolution of finite subordination in the Uralic languages, or the cross-linguistically attested development of phrases in Spec,CP into heads in C) support the following chronological order:

At stage 1 of the grammaticalization path, hogy is a phrasal expression with lexical (manner) content in a paratactic clause. At stage 2, it also assumes a subordinating feature, i.e., it is a [+C, +Op] expression moved to Spec,CP, turning its sentence into a free relative. At stage 3, it loses its lexical content and its phrasal category; it does not bind a variable anymore; it is simplified into a [+C] head base-generated in C.

The evolutionary path outlined in (37) is not specific to Hungarian; the starting point of the grammaticalization of the general complementizer has been claimed to be the same correlative construction in various Indo-European languages. The difference between the evolution of the Hungarian hogy and the Germanic that-type complementizers may stem from a minor difference in the selectional properties of Hungarian and Germanic verbs of saying: whereas in Hungarian they were intransitive, with an optional oblique complement, in the Germanic languages they were transitive. Consequently, if the evolution of the Germanic complementizer followed a path similar to that of hogy, it started out as the relative pronominal correlate of a demonstrative object, whereas hogy was originally the correlate of a demonstrative manner adverb. This difference disappeared when the relative pronoun/proadverb ceased to bind a variable with semantic features and came to be reanalyzed as a complementizer head base-generated in C. In Hungarian, complement clauses developed a pronominal associate so as to satisfy their need to be represented in the focus or topic slot of the main clause, and to satisfy the morphological case requirement of the matrix predicate. In English, by contrast, the disappearance of morphological cases and the rigidity of word order led to the redundancy and the eventual disappearance of the pronominal associate.