Reconstructing the decoupling of case and agreement in Old Hungarian: Evidence from epithets and names as syntactic fossils 1

The interdependence of accusative case and object agreement has changed dramatically during the history of Ugric languages. While Proto-Ugric exhibited full interdependence (mediated by topicality), this connection has loosened in the extant Ob-Ugric languages (Mansi and Khanty) and it is severed completely in Late to Modern Hungarian. In this paper, I introduce new, hitherto unreported empirical evidence (from nicknames and family names that preserve archaic syntactic features) for an intermediate stage of Early Old Hungarian (which predates our earliest written records) where case assignment was still a function of topicality but object agreement was already a function of definiteness. In addition to providing insight into an unrecorded stage of Hungarian, my findings also contribute to a more thorough understanding of the connection between case, agreement and information structure in Ugric and beyond.


Introduction
According to Baker (2015) case and agreement can be (A) interdependent, (B) partially independent or (C) independent.É. Kiss (2020) argued that within the Ugric family, extinct Eastern Mansi and reconstructed Proto-Ugric exemplified (A), the Ob-Ugric languages of today (Mansi and Khanty) display (B) and Modern Hungarian is an example of (C).In Eastern Mansi, only topical objects elicited verbal agreement (so-called object agreement) and only topical objects received accusative case.In the other Ob-Ugric variants, while verbal agreement is still 1 I would like to thank Gergely Fórizs, Katalin É. Kiss, František Martínek, Ádám Nádasdy, György Rákosi, Péter Rebrus and Radek Šimík and the members of the research group "Implications of endangered Uralic languages for syntactic theory and the history of Hungarian" for comments and advice.I would like to express my gratitude to Ferenc Vörös for graciously sharing his database of Hungarian surnames.I am thankful to the audiences of various talk versions of this paper at presented at ICHL 25, DIGS 23 and within the Hungarian Research Centre for Linguistics.I am indebted to the editors and reviewers of Diachronica for helpful comments and advice.The research reported here was supported by Grant 129921 of the National Research, Development and Innovation Fund of Hungary, Grant TKP2020-NKA-11 of the Ministry for Innovation and Technology of Hungary, the Bolyai scholarship of the Hungarian Academy of Sciences and the Bolyai Plus scholarship of the New National Excellence Programme of the Ministry of Innovation and Technology of Hungary.I am grateful to the Budapest Library of Central European University for graciously providing me a place to work during the winter of 2022-2023.adjective or noun without any morphological marking.This is mostly used to create slurs or nicknames based on a characteristic trait: a typical and defining attitude, disposition, or activity: (1) know-nothing 'does not know anything' -> 'ignoramus' (English) (2) vau-rien 'is worth nothing' -> 'useless person' (French) Such epithets often develop into family names and are preserved as such: (3) Shakespeare, Makepeace, Drinkwater (English) (4) Boileau (= boit l'eau drinks the water -> teetotaler) (French) Modern Hungarian has a set of such epithets/names which preserve a peculiar syntactic pattern: a non-casemarked object followed by a verb carrying the object agreement suffix: 3 (5) hús-(nem)-esz-Ø-i wine-not-drink-3SG-OBJ 'wine-lover/teetotaler', lit.'drinks (not) wine' (6) ló-dönt-Ø 4 -i horse-topple-3SG-OBJ 'strong enough to topple a horse', lit.: 'topples horse' This pattern is unexpected as in Modern Hungarian, objects are obligatorily overtly accusativemarked independently of their informational structural status (topical or otherwise) or their definiteness.As I will argue below, these fossils represent an earlier stage of Hungarian where accusative case assignment was still a function of topicality (with only topicalized objects receiving accusative case, as is the case in Proto-Ugric and all the Ob-Ugric varieties), but object agreement was already a function of definiteness and not of topicality (similarly to the situation in Late Old to Modern Hungarian).This suggests that Hungarian traversed a different path than its Ob-Ugric sisters: the agreement-topicality link was severed first and the case assignment-topicality link was severed later: 3 This strategy is not limited to transitive verbs with an overt object.Sentences (i) without an object (ii) or with a silent pro object (Hungarian being a pro-drop language) have also been turned into epithets (cf. the Appendix): (i) jö-tt-Ø-men-t-Ø come-PST-3SG-go-PST-3SG '(he/she) came and went' -> 'rootless newcomer, carpetbagger' (ii) pro-hány-Ø-ja-vet-Ø-i it-scatter-3SG-OBJ-throw-3SG-OBJ 'lobs it around and scatters it wide' -> 'careless, negligent' 4 The 3SG subject agreement suffix is phonologically null (silent) in Hungarian.All the other subject agreement suffixes have are non-silent.Old to Modern Hungarian is an example of total independence.
The link connecting case assignment and object agreement in Ugric is information structure.
In Eastern Mansi (Kulonen 1989, Virtanen 2012, 2015), the most conservative (and by now, extinct) Ob-Ugric dialect, there is a complete fusion of topicality and argument hierarchy.The word order is strictly SOV.Subjects obligatorily function as the so-called primary topic of the clause (so much so that in case a subject is not appropriately topical, being non-referential or non-specific, obligatorily passivisation occurs involving the demotion of the subject to a byphrase and the promotion of a referential or specific object to subject position).Discourse-old objects are accusative-marked and elicit object-verb agreement (7a), whereas discourse-new objects are uncasemarked and do not elicit object-verb agreement (7b): (7 ) a. pro õõw-mø öät kont-iil-øm.É. Kiss (2020) provides an account for the case-marking properties in terms of dependent case theory (Marantz 1991, Baker 2005), making the crucial assumption that vP and SubjP represent two phases 5 , i.e., two separate spellout domains for case assignment.In case the object is topicalized, it ends up in a single domain with the subject (SubjP, meaning the span from V to SubjP, since vP, being a phase, has already been sent to spellout and is thus, in essence, invisible for further syntactic operations): since nominative case goes to the subject, the 5 According to standard generative assumptions, the derivation of a clause proceeds in a bottom-up fashion and by phases: roughly speaking, a phase is a chunk of syntactic structure that is considered closed (impenetrable) for further syntactic operations: once a so-called phase head (such as v) has been inserted, the sister (complement) of that head becomes impenetrable: no element can leave it by movement and its contents become inaccessible for operations involving an element external to the phase (Chomsky 2001).
topicalized objects receives accusative case. 6If the object remains in situ, it ends up being the only NP within its spellout domain (vP) and thus, it receives nominative case.
In other, more innovative and non-extinct variants of Ob-Ugric (Nikolaeva 1999, Csepregi 1997, 2019, Asztalos, Gugán & Mus 2017, Riese 2001, Skribnik 2001, Sosa 2017, Filchenko 2007, Bíró & Sipőcz 2017), verbal agreement is still a function of topicality.However, the link between accusative case assignment and topicality is severed.All objects are assigned structural Non-topical pronouns emerge in contrastive contexts, and they fail to elicit verbal agreement (9b).Independently from topicality, however, the pronominal objects are structurally and morphologically accusative-marked.To account for these patterns, É. Kiss (2020) argues that the innovative Ob-Ugric dialects have the same basic syntactic structure as Eastern Mansi (and Proto-Ugric), with the difference that vP and SubjP are not separate phases, i.e., they do not form independent spellout domains.This means that independently of topicality, the object is always in the same domain as the subject (and c-commanded by it), and as a result, it always receives accusative case.
In contrast to its Ob-Ugric sister languages, Modern Hungarian is no longer SOV and it has a richly articulated discourse-functional left periphery, with dedicated and movement-derived topic and focus positions: Mary the book-ACC PRT-read-PAST-OBJ-3SG 'As for the book, Mary read it.'(topicalized, definite object) f.Mari egy könyv-et el-olvas-ott-Ø.

Mary a book-ACC PRT-read-PAST-3SG
As can be seen in ( 11), in Modern Hungarian, all objects are assigned structural and morphological accusative case, and object agreement on the verb is triggered by definite objects (independently of discourse function): thus, both agreement and case assignment are fully independent from discourse function.
É. Kiss (2020) argued that the gradual separation of agreement and case from topicality was due to the loosening of the strict SOV structure of Proto-Ugric: the increasing frequency of post-verbal objects in Ob-Ugric (SVO) and the development of a discourse functional left periphery (Top Foc V X*) in Hungarian.As far as Ob-Ugric languages (dialects of Mansi and Khanty) are concerned, the diachronic changes have been mapped in great detail (É.Kiss 2020).
Hungarian, however, has proved to be more of a challenge since Late Old Hungarian (the earliest period from which we have surviving texts, 12th C) had the same system in terms of accusative case assignment and object agreement as Modern Hungarian.It is only through some fossilized constructions that were still detectable in Late Old to Middle Hungarian that we have been able to gain some insight into the case and agreement system of undocumented Early Old Hungarian.
The or that objects in general were assigned accusative case, the exponent of which in the case of lexical nouns was a phonologically null accusative morpheme (as in Eastern Khanty, Northern Khanty and Northern Mansi).The fact that in (17), the reflexive pronoun has no visible case marking supports the former position.This suggests that Hungarian traversed a different path than its Ob-Ugric sisters.In Hungarian, the agreement-topicality link was severed first and the case assignment-topicality link was severed later; whereas in Ob-Ugric, the case assignmenttopicality link was severed first, and the agreement-topicality link is still intact (cf.Table 1 above).
12 When the object is lexically specified as indefinite (such as the universal quantifier minden), or there is no object at all, we witness a lack of object agreement (OBJ), as expected: (i) minden-tud-Ø everything-know-3SG 'knows everything' -> 'know-all' (ii) ingyér-ád-Ø for.free-give-3SG'gives without asking for anything in return' -> 'munificent, generous'

Analysis
In this chapter, I first present my proposal: I argue that Early Old Hungarian was conservative in having a Proto-Ugric-like syntax where the minimal verb phrase formed a separate case assignment domain; however, it was also innovative in that object agreement was already sensitive to definiteness (as opposed to topicality).Afterwards, I discuss two potential challenges to my proposal: I will argue against the possibility of an incorporation analysis and provide an account for the underrepresentation of the -ja allomorph of the OBJ object agreement suffix.

A dependent case analysis of Early Old Hungarian
To recapitulate, É. Kiss (2020) proposed the following sentence structure for (conservative, strictly SOV) Ob-Ugric (see (8) above): Non-topical objects remain in-situ inside the minimal verb phrase (VP), whereas topicalized objects move to a position external to the extend verb phrase (vP).Furthermore, proposing a dependent-case style analysis (Baker 2015), she argued that the interdependence or otherwise of accusative case assignment and topicality is a function of whether SubjP and vP are separate domains for case assignment.If they are (in other words, if v is a hard phase head), then nontopical objects, being in a separate case assignment domain from the subject, receive nominative case, whereas topical objects, being in the same case assignment domain as the subject, receive accusative case (Proto-Ugric, Eastern-Mansi).This is shown below: If v is a soft phase head then the whole Subject Phrase (SubjP) forms a single case assignment domain.In this case, only the subject receives nominative case and the object receives accusative case independently of whether it has been topicalized or remained in situ (Eastern and Northern Khanty, Northern Mansi, Late Old to Modern Hungarian): I argue that in the stage of Early Old Hungarian that is preserved in the nickname construction under discussion, the sentence still had the structure shown in (24), and v was still a hard phase head (as it was in Proto-Ugric and Eastern Mansi): non-topical objects, being in a separate case assignment domain from the subject, received nominative case (and thus no visible case marking, as nominative is morphologically unmarked in Hungarian).
Object agreement, on the other hand, was already sensitive to the  1975:158) and it is a development that is orthogonal to whether v is a soft or a hard phase head.
In other words, the case-agreement-topicality link can, in effect, be broken down into two independent links: the case-topicality link and the agreement-topicality link.Since these two are independent, any one of the two can be severed independently of the other: there is no 'natural order' in which the severing of the two links is supposed to happen.And indeed, both of the logically possible orders have been instantiated diachronically: by Ob-Ugric (Mansi and Khanty) on the one hand, and Hungarian on the other.

Against an incorporation analysis
Before concluding, there are two potential challenges to our analysis that need to be addressed.
The first one concerns a potential alternative analysis.Since the objects in the construction under investigation are non-case-marked and they directly precede the verb (unless a negator intervenes), one might be inclined to propose an object incorporation analysis.However, such an analysis is manifestly unfeasible for a variety of reasons.Consider first that incorporated objects in Hungarian fail to elicit object agreement (unlike the objects in our construction): (26) a. Mari meccs-et néz-Ø.'drinker of water only'

The underrepresentation of the -ja allomorph
In Late Old to Modern Hungarian, the object agreement suffix (OBJ) is subject to allomorphy conditioned by (i) the subject agreement suffix and (ii) by the vowel quality of the verbal stem.
In case the subject agreement suffix is 3SG (phonologically null), OBJ has two allomorphs in standard Modern Hungarian: -ja (which attaches to stems with back vowels) and -i (which attaches to stems with front vowels): 16  1913, Horger 1931, Imre 1971): this means that this pattern, while irregular in the standard dialect, counts as regular in many non-standard dialects.
The second reason is that a strong analogical effect might have helped the survival (and indeed predominance) of the -i forms in epithets: namely, the existence of an unrelated but homophonous adjectivizing -i suffix which also happens to be used in epithets: (31) a. csoszog-i shuffle.feetV-ADJ'foot-dragger' b. ki-vagy-i who-be.2SG-ADJ'pretentious' c. vigyor-i grinN-ADJ 'prone to an easy smile' This -i suffix is clearly different from both the -i (OBJ) suffix and the standard diminutive suffix (which also happens to have the phonologial form of -i).This epithet-i is omnivorous in the sense that it can attach to verb stems (31a), conjugated verb forms (31b) or even nouns (31c).
It is non-harmonic, having a single allomorph (-i).Unlike the standard diminutive -i, it is nontemplatic (the standard diminutive -i always has a two-syllable output: aranyos 'cute' -> ari 'cute-DIM').And, unlike the standard diminutive, it typically induces a change of word class, turning whatever input it receives into an adjective. 17But, crucially for our purposes, it is an -i suffix that features prominently in epithets, and as such, it may have played a reinforcing role by way of analogy, helping the spread of the dialectally available non-harmonic form (back stemfront vowel).

Conclusion
In this paper, I examined a peculiar syntactic pattern of Early Old Hungarian that was fossilizied and preserved in certain epithets and names: a non-casemarked object followed by a verb displaying object agreement.I argued that these fossils represent crucial new evidence which helps us to reconstruct how case assignment and object agreement worked in Early Old Hungarian, a stage of Hungarian of which we have no written records whatsoever.I have shown that Early Old Hungarian represented a transitional stage between Proto-Ugric (where both case assignment and object agreement depended on topicality) and Late Old to Modern Hungarian (where both case assignment and object agreement are independent of topicality).In Early Old Hungarian, case assignment was still a function of topicality but object agreement was already a function of definiteness.This is a pattern that differs from what we have seen in other Ugric languages, where the agreement-topicality link was severed first and the case-topicality link 17 I am thankful to Péter Rebrus for discussion on the morphophonology of this diminutive suffix.
only later.These findings thus fill a gap in the history of Hungarian (and Ugric) and demonstrate the independence of the agreement-topicality link and the case-topicality link.
1SG door-ACC NEG find-SG-1SG 'I don't find the door.' (Eastern Mansi, Virtanen 2014: 405) b.Kom jowt-nyõõl wø-s.man bow-arrow take-PST.3SG'The man took a bow and an arrow.' (Eastern Mansi, Virtanen 2014: 407) É. Kiss (2020) argues that discourse-old objects are extracted from the verb phrase (VP) and moved to a so-called secondary topic position, where they enter into an agreement relation with the verb and are assigned accusative case; whereas discourse-new objects remain in situ within the VP, do not enter into an agreement relation and fail to receive accusative case.Considering these facts and other evidence (such as adjunct placement patterns), É. Kiss (2020) proposes the following structure for Eastern Mansi (and reconstructed Proto-Ugric): generated in the specifier of vP (t1) and moved to the specifier of SubjP (NP1).Discourse-old (topical) objects are base-generated in the speficier of VP (t2) and moved to the specifier position of ObjP (NP2).Discourse-new objects are base-generated in t2 and remain there.The heads of the subject phrase and the object phrase (Subj and Obj) are spelled out as subject and object agreement suffixes.
thus be found in one of three positions.Discourse-functionally neutral objects remain in situ (within the vP), in a post-verbal position; focused objects are preverbal and induce particle-verb inversion; topical objects are also preverbal but they do not induce particleverb inversion.Consider: (11) a. Mari el-olvas-t-a-Ø 7 a könyv-et.Mary PRT-read-PAST-OBJ-3SG the book-ACC 'Mary read the book.' (discourse functionally neutral, definite object) b.Mari el-olvas-ott-Ø egy könyv-et.Mary PRT-read-PAST-3SG a book-ACC 'Mary read a book.' (discourse functionally neutral, indefinite object) c.Mari a könyv-et olvas-t-a-Ø el.Mary the book-ACC read-PAST-OBJ-3SG PRT 'It was the book that Mary read.' (focus-moved, definite object) d.Mari egy könyv-et olvas-ott-Ø el.Mary a book-ACC read-PAST-3SG PRT 'It was a book that Mary read.' (focus-moved, indefinite object) e. Mari a könyv-et el-olvas-t-a-Ø.

Table 1 . Topicality, agreement, and accusative case assignment in the Ugric Languages Proto-Ugric, E Mansi E & N Khanty, N Mansi Reconstr. Early Old Hungarian Modern Hungarian
Ugric and the Ob-Ugric languages.However, as far as object agreement was concerned, Early Old Hungarian was more innovative: in a departure from Proto-Ugric and the known Ob-Ugric varieties, object agreement was already sensitive to the [+definite] feature, as opposed to [+topic] Such a change from [+topic]-sensitivity to [+definite]-sensitivity is a crosslinguistically well-attested phenomenon (facilitated by the shared component of specificity/givenness, cf.Givón 1975:158) and it is a development that is orthogonal to whether v is a soft or a hard phase head.The paper is organized as follows.In Chapter 2, I discuss the landscape of case and agreement in Ugric and the challenge of Hungarian.In Chapter 3, I present the new evidence: archaic syntax preserved in names and epithets.In

Archaic syntax preserved in personal names
Happy is the woman who is bought together with her child.'(Literally: 'Happy is the woman whom they buy together with her child.')more relevant common characteristic of these sentences is that they provide a general characterization of the subject's attitude/disposition/relationship with regard to the object: whether she eats meat / drinks wine / topples horses in general.In other words, the object is generically interpreted.Crucially, in Hungarian, singular definite DPs can freely receive a constructions include SOV 8 non-finite participial embedded clauses with non-casemarked (or nominal-marked) 9 objects in Late Old Hungarian: 'Him having thought this, the Lord's angel appeared to him.' Munich Codex (1416/1466): p. 8 verso; cited by É.Kiss (2020:417)The other relevant fossilized construction is variable object agreement with topical indefinite objects.While in Modern Hungarian, indefinite objects fail to trigger verbal object agreement, Personal names (such as nicknames or nickname-derived family names) often preserve archaic features of phonology, morphology, and even syntax (for the latter, cf.Layton 1990 and Bowern 1998).Of peculiar interest for our purposes is the cross-linguistically well-attested strategy of turning a clause-sized element into an adjective or noun without any morphological marking.This is mostly used to create epithets or nicknames based on a characteristic trait: a typical and defining attitude, disposition, or activity.Such constructions are attested in a variety of languages (see examples in the introduction).Modern Hungarian has a set of such epithets/names 10 which preserve a peculiar syntactic pattern: a non-casemarked object followed by a verb carrying the object agreement suffix:(14) bor-(nem)-isz-Ø-sza the grandchild.1SGnoteat-3SG-OBJPRT the meat.ACC 'My grandchild does not eat any meat.'Thisleavestheoptionthat object agreement here is already sensitive to definiteness (as it is in Late Old Hungarian).The question at this point is whether it is reasonable to assume that the object in sentences such as (18) is indeed definite.Note first that the reflexive pronoun maga 'self' is indisputably definite (20) in Late Old to Modern Hungarian: A 'Mary likes wine.' Thus it is reasonable to assume that the generically interpreted object in (20) was indeed definite in Early Old Hungarian too (cf.Egedi 2013:378 for a detailed argument), and object agreement on the verb was indeed triggered by definiteness.12Thesefossils thus arguably represent a stage where object agreement was already a function of definiteness (and not of topicality) and non-topicalized lexical noun phrase objects were morphologically non-casemarked.This latter fact may either indicate that non-topicalized objects were not assigned accusative case (as in Eastern Mansi and reconstructed Proto-Ugric)

Table 2 : Stem-OBJ patterns
Berzsenyi was a wine-avoider, or rather, someone who drank only water.'16For a detailed discussion of vowel harmony in Hungarian, cf.Rebrus & Törkenczy 2015 and references.