From narrative past to mirativity and direct evidentiality: the case of Moldavian (Csángó)

: This paper describes the evolution of grammaticalized evidentiality in the Moldavian dialect of Hungarian. It documents how the su ﬃ x -a/e , originally the marker of narrative past, became a rare, elevated marker of past tense highlighting signi ﬁ cant past events; how it assumed a mirative overtone; and how the features ʻ witnessed ʼ and ʻ immediate past ʼ , often present in mirative utterances, became inherent parts of its meaning. This grammaticalization path has resulted in an evidential system with typologically unique features. It is a two-term system based on the opposition of direct evidentiality and no evidentiality – violating the alleged universal that if a language has grammaticalized direct evidentiality, it has also grammaticalized indirect evidentiality. Mirative meaning is expressed by the same - a/e su ﬃ x that also encodes direct evidentiality – whereas it is claimed to be the extension of inferred evidentiality elsewhere. The unique properties of Moldavian Hungarian evidentiality are derived from the historical evolution of the - a/e su ﬃ x.


Introduction
This paper points out an evidential system in the Moldavian (Csángó) dialect of Hungarian that is at variance with the typological characterization of grammaticalized evidentiality (Aikhenvald 2003(Aikhenvald , 2004(Aikhenvald , 2012De Haan 2012;Lazard 2001, etc.) in the following respects: (i) Moldavian Hungarian has past tenses not referring to the source of information, and past tenses (perfect immediate past and imperfect past) expressing direct evidentialitythereby violing the alleged universal that if a language has grammaticalized direct evidentiality, it has also grammaticalized indirect evidentiality.
(ii) In Moldavian Hungarian, mirative meaning is expressed by the same -a/e suffix that also encodes direct evidentialitywhereas it is overwhelmingly associated with the marker of inferred evidentiality elsewhere. (iii) Mirativity preceded direct evidentiality in the functional evolution of the -a/e suffixas opposed to the semantic paths reconstructed for other languages (Aikhenvald 2004(Aikhenvald , 2012.
These properties will be shown to be consequences of the historical evolution of the -a/e suffix. The analysis of Moldavian Hungarian evidentiality describes a phenomenon that is absent from other versions of Hungarian and is not taken notice of by Hungarian grammars. It also contributes to the typological classification of evidential systems by introducing a rare new type, and furthermore, it shows that the correlation hypothesized between mirativity and indirect evidentiality in two-member evidential systems actually holds between mirativity and the evidentially marked form, irrespective of whether it encodes direct or indirect evidentiality. On a more general level, the analysis of the evolution of Moldavian Hungarian evidentiality demonstrates how the pragmatic conditions of the use of a grammatical form can grammaticalize into inherent meaning components.
The two-term evidential system of Moldavian Hungarian, contrasting the -a/emarked past tenses expressing direct witnessing with the evidentially neutral -t-marked past tenses, is introduced in Section 2. Section 3.1 surveys the functional evolution of the -a/e suffix and traces back its history to Old Hungarian, where it was the marker of narrative past, the suffix of past events unrelated to the present. Section 3.2 shows how the -a/e suffix gradually lost ground to -t, the former perfect marker and how it assumed, because of its rarity, an emotionally loaded mirative overtone in the Middle Hungarian period. Section 3.3 demonstrates the reinterpretation of -a/e from the marker of mirative past to a suffix of witnessed mirative past and witnessed immediate past in Modern Hungarian. Section 4 revisits the atypical properties of evidentiality and mirativity marking in Moldavian Hungarian and derives them from the developmental path of -a/e. Section 5 is the conclusion. their sixties, one of them a native of the village Cleja/Klézse, the other a native of Faraoani/Forrófalva. Hungarian is still fully functional in these villages, and our consultants are Hungarian-dominant bilinguals. The consultant from Forrófalva only speaks the Moldavian version of Hungarian. The consultant from Klézse was also exposed to the vernacular Hungarian of the Transylvania region of Romania for some time, and he often made comments on the differences of the Transylvanian standard and his own dialect. Our consultations involved grammaticality judgement tasks, evaluations of minimal pairs, and discussions of the (in)appropriateness of sentences with different past tenses in various contexts. Two sources covering all three Moldavian Hungarian subdialects have also been consulted: the Moldavian Hungarian database of the Hungarian Research Centre for Linguistics (https://nlp.nytud.hu/csango/), containing video recordings and their transcriptions recorded by Vilmos Tánczos in the 1990s, and Péntek's (2018) dictionary of Moldavian Hungarian, containing text examples from archived materials of Moldavian fieldwork carried out by the Hungarian Linguistics Department of Babeş-Bolyai University in the second half of the 20th century. Their data indicate that the evidential system outlined below must be, or must have been, present in all three subdialects of Moldavian Hungarian.

The tenses system of Moldavian Hungarian
Whereas Standard Hungarian only has a single past tense (and an unmarked non-past), Moldavian Hungarian has preserved the complex tense system of Old Hungarian. It has five tenses, which are labelled in terms of the Hungarian grammatical tradition as follows: (1) The Moldavian Hungarian tense system (based on Hungarian grammatical tradition) i. present V+AGR mond-om say-1SG ʻI sayʼ ii. general past: V+t+AGR mond-t-am say-PST-1SG ʻI saidʼ iii. past perfect: V+t+AGR AUX+t mond-t-am vó-t 4 say-PST-1SG be-PST ʻI had saidʼ iv. immediate past: V+a/e+AGR mond-á-m say-APST 5 -1SG ʻI have just saidʼ 4 The past perfect tense consists of a lexical verb bearing tense and agreement suffixes, and a temporal auxiliary (the be-verb) bearing a tense suffix. The tense suffix on the lexical verb was originally an aspect marker. 5 APST stands for -a/e-marked past.
v. imperfect past: V+AGR AUX+a mond-om val-a say-1SG be-APST ʻI have been saying; I was sayingʼ As will be argued below, two of the four past tenses have developed evidential and mirative functions. Whereas the tenses marked by -t, i.e., the general past (2.2.1) and the past perfect (2.2.2), are evidentially neutral, the tenses marked by -a/e, i.e., the immediate past (2.2.3) and the imperfect past (2.2.4), encode direct evidentiality.

General past
The suffix -t is the general past-tense suffix in Moldavian Hungarian as it is in Standard Modern Hungarian. The -t-marked past can be used to describe any situation that precedes the speech time. It is the tense of narrativesirrespective of whether or not the event was witnessed by the speaker (2a-b). It can be used in the case of past events with a present result (2c), recent past events (2d), extended past events (2e), past events prior to a past reference point (2f), etc. It can occur with time adverbials of all types.

Past perfect
The verbal complex consisting of a lexical verb bearing the -t past-tense suffix and an auxiliary (the be-verb) also bearing the -t suffix expresses past perfect tense. The agreement suffix cross-referencing the subject and the definite object (if any) is carried by the lexical verb. The two instances of -t iconically mark past events that precede a past reference point, which is often determined by the tense of a superordinate or a coordinate clause: (3) a. El-te-tt-em a level-et, ami-t a fi-am küld-ött vó-t.
PRT-keep-PST-1SG the letter-ACC which-ACC the son-1SG send-PST.3SG be-PST ʻI kept the letter that my son had sent to me.ʼ b. Ve-tt-ünk vó-t tavaly malac-ot, s az meg-dögl-ött. buy-PST-1PL be-PST last.year piglet-ACC and it  ʻWe had bought a piglet last year and it died.ʼ

Immediate past
Verbs supplied with the -a/e past tense suffix (glossed as APST) denote recent events, in most cases events that barely preceded the speech time. Verbs marked by -a/e can cooccur with adverbials like most ʻnowʼ, az imént ʻjustʼ, tíz minutja ʻten minutes agoʼ, but they may refer to the immediate past also without any time adverbial: (4) a. Ki-ereszt-é-m a tyuk-ok-at. out-let-APST-1SG the chicken-PL-ACC ʻI have (just) let out the chickens.ʼ b. Az imént ev-é-m /ev-é-nk. the now eat-APST-1SG /eat-APST-1PL ʻI/We have just eaten.ʼ c. Most jöv-é-nk bé az ógradá-ból. now come-APST-1PL in the courtyard-from ʻWe have just come in from the courtyard.ʼ d. Tíz minut-ja, hogy el-kel-é-k az asztal-tól. ten minute-3SG that off-get-APST-1SG the table-from ʻIt is ten minutes since I got up from the table.ʼ The attested examples involving an -a/e-marked verb all refer to events immediately preceding the speech time, but our consultants also accepted examples like (5a), representing recent past rather than immediate past. However (5b), referring to an event preceding the speech time by several weeks, was rejected.
(5) a. Tennap hamar kell-e fel-kel-j-ek. yesterday early need-APST.3SG up-get-SBJV-1SG ʻYesterday I needed to get up early.ʼ b. *Nyárba meg-jár-á-nk a fiu-nk-at Budapest-en. summer.in PRT-visit-APST-1PL the son-1PL-ACC Budapest-in ʻIn the summer we visited our son in Budapest.ʼ A further condition of the use of the -a/e past is that the denoted event be completed before the speech time. That is, the suffix marks perfect immediate past, as revealed by the interpretation of example (6):

(6)
A gyermek bele-es-é-k a gödör-be. 6 the child into-fall-APST-3SG the pit-into ʻThe child fell into the pit.ʼ As our consultants explained, this sentence means that the child fell into the pit just now but he is not in there any more; he has been taken out. (Apparently, a past action with an ongoing result state is not regared as completed by our consultants.) The most conspicuous condition of the use of the -a/e past is that the given activity be carried out by the speaker himself/herself, or be experienced by him/her via seeing, hearing or smelling: (7) a. Most ki-lel-é-m, mi-t akar-sz. now out-find-APST-1SG what-ACC want-2SG ʻNow I have found out what you want.ʼ b. "Hol a gyermek?" "Lát-á-m az imént az ógradá-ba." where the child see-APST-1SG the now the courtyard-in "Where is the child?" "I have just seen him/her in the courtyard." c. Men-é-k itt fel, sérz-é-m a hajma-bűz-t. go-APST-1SG here up and feel-APST-1SG the onion-smell-ACC ʻI have passed here and I have felt the smell of onion.ʼ d. Hall-á-m a durrogás-t. hear-APST-1SG the thunder-ACC ʻI have heard the thunder.ʼ The subject of the -a/e-marked verb is usually 1st person singular or plural. 2nd and 3rd person forms mostly occur in dialogues as the speakerʼs reactions to a previous question or statement (8a), or in mirative exclamations (8b): The -a/e suffix cannot be used if the event in the immediate past was not personally witnessed by the speaker. Thus the following utterancessaid with the school and the church out of sightare judged to be unacceptable: (9) a. %Dél van, most indul-á-nak haza a gyermekek az eskolá-ból.
noon is now start-APST-3PL home the children the school-from ʻIt is noon; the children have just set off home from school.ʼ b. %Tíz minut-ja, hogy harangoz-nak val-a; már meg-kezdül-e ten minute-3SG that ring-3PL be-APST already start-APST.3SG a mise. the mess ʻIt is ten minutes since they were ringing [the bell]; the mess has already started.ʼ The -a/e past can also be used in sentences expressing surprise, and anger, annoyance or joy caused by an unexpected event, i.e., it can also encode mirativity. Of the mirative values distinguished by Aikhenvald (2012), we find sudden discovery, surprise, and unprepared mind in the attested examples. Our consultants often began sentences of this type with Istenem! ʻMy God!ʼ, or Uram Istenem! ʻMy Lord, my God!ʼ. They said that such sentences are addressed to themselves.
(10) a. Essze-tör-é-m a mesin-om-at! together-break-APST-1SG the car-1SG-ACC ʻI have damaged my car!ʼ b. Isten-em, az a bolond essze-tör-é a mesin-om-at! God-1SG that the fool together-break-APST.3SG the car-1SG-ACC ʻMy God, that fool has damaged my car!ʼ c. Ur-am Isten-em, el-ver-é a jég a puj-t! Lord-1SG God-1SG up-cut-APST.3SG the hail the corn-ACC ʻMy Lord, my God, the hail has cut up the corn!ʼ d. Isten-em, nyer-é-k 100 frank-ot! God-1SG win-APST-1SG 100 franc-ACC ʻMy God, I have won 100 francs!ʼ 7 In mirative exclamations of this type, the recency condition is not always observed. The requirement of witnessing, however, holds in this case, too; if it is not satisfied, the -a/e past is rejected: The -a/e past is rare in negated sentences, and even rarer in questions (12a-b). It is ungrammatical in if-clauses (12c), where the requirement of immediate precedence with respect to the speech time is hard to satisfy.
The -a/e past is seldom found in the narrative documents of Moldavian Hungarian, among them life stories, tales, and ethnographical explanations (see, e.g., Rubinyi 1901Rubinyi , 1902Rubinyi , 1905Hegedűs 1952); they are more common in conversations recorded by field workers. The reason must be that this tense is used to describe events that take place in the speech situation in which the given utterance arises.

Imperfect past
The imperfect past is a complex verb form, involving a lexical verb supplied with an agreement suffix, and the temporal auxiliary val-a (cognate with the be-verb) carrying the -a/e past-tense suffix (see Section 2.2). The -a/e suffix encodes witnessed past in this case, too. Whereas the -a/e-marked simple verb discussed in Section 2.2.3 expresses witnessed perfect immediate past, the V+AGR val+a form encodes witnessed imperfect past, often witnessed imperfect immediate past.
The participation or witnessing requirement on the use of the -a/e suffix cannot be lifted, but the immediacy condition is not always observed. Thus, V+AGR val+a can be used to describe continuous immediate past events (13a-c), and repeated or habitual past events (14a-d). In most cases, one of the arguments (the implicit goal in [14b]) is the 1st person pronoun.

Interim conclusion
The notions of evidentiality and mirativity are not parts of the Hungarian grammatical tradition, but the above observations indicate that they play a crucial role in the grammar of Moldavian Hungarian. Consequently, the terminology describing the Moldavian Hungarian tense system is to be revised as follows: (16) The tense system of Moldavian Hungarian i. present V+AGR mond-om say-1SG ʻI sayʼ ii. general past: V+t+AGR mond-t-am say-PST-1SG ʻI saidʼ iii. past perfect: V+t+AGR AUX+t mond-t-am vó-t say-PST-1SG be-PST ʻI had saidʼ iv. witnessed (mirative) immediate past: V+a/e+AGR mond-á-m say-APST-1SG ʻI have just saidʼ v. witnessed imperfect past: V+AGR AUX+a mond-om val-a say-1SG be-APST ʻI have been/I was sayingʼ

Atypical properties of Moldavian Hungarian evidentiality
The grammatical encoding of evidentiality is common in the languages of the world, and linguistic typology has extensively described, compared and classified the general and diverse properties of the different evidential systems (see, e.g., Aikhenvald 2004Aikhenvald , 2018. A criterion of classification is whether evidentiality is expressed autonomously or fused with a grammatical category, and in the latter case, what grammatical category it is fused with (Aikhenvald 2015). The situation found in Moldavian Hungarian, where evidentiality marking is restricted to the past tenses, and the evidentiality marker is fused with a past tense suffix, is common crosslinguisticallysee also Visser (2015) and Forker (2018).
Moldavian Hungarian represents a further type: A6. Firsthand versus everything else.
Type A6 is predicted in the literature to be impossible, albeit the reasoning, based on intuition, is far from being conclusive. According to de Haan (2012: 11-12), it would seem to be a near absolute universal that if a language has grammaticalized direct evidentials, it also has grammaticalized indirect evidentials. Or, to put it another way, indirect evidentials are grammaticalized before direct evidentials. Pragmatically, the reasoning seems to be that statements made by the speaker are assumed to convey direct evidence unless stated otherwise. Hence the abundance of lexical material and grammaticalized evidentials to mark indirect evidence, as this is the marked member of the opposition. Once grammaticalized indirect evidentials are present, direct evidentials can be grammaticalized.
DeLancey (2012: 544-545) makes a similar claim, arguing that the particle lõ in the Athabaskan language Hare cannot be a direct evidentiality marker because the language does not mark indirect evidentiality, and "one cannot coherently describe a hypothetical language in which there was a marked construction indicating ʻdirect sensory evidenceʼ contrasting only with an unmarked construction with no evidential value at all". Hill (2012) rejects this argument; for him it seems a perfectly legitimate strategy for speakers not to mark the source of information unless it has been obtained from direct personal experience. That is, Hill does not exclude the possibility that Hare represents a formerly unattested type, the one identified above as A6.
A further unusual feature of Moldavian Hungarian is that mirativity is expressed by the suffix encoding direct evidentiality. Mirativity has been claimed to be an extension of indirect evidentialityunless mirativity marking is a system independent of evidentiality, encoded by a separate morpheme, in which case it can cooccur with any type of evidentiality (Aikhenvald 2004(Aikhenvald , 2012DeLancey 1997DeLancey , 2001. Mirativity, along with hearsay and inferential, is a use of the non-firsthand, or mediative, evidential in Lazardʼs (2001) description of evidentiality in Balkan and Middle Eastern languages, as well. Witnessed evidentials with mirative overtones have been reported from two languages: Cupeño (Hill 2005) and Lhasa Tibetan (DeLancey 1997(DeLancey , 2012, but Aikhenvald (2012: 467) found these examples inconclusive. The problem with Hillʼs analysis of Cupeño is that when she discovered mirativity in texts and notes collected by her decades earlier, there were no more Cupeño speakers left, hence there was no way to examine mirativity with targeted tests. DeLanceyʼs claim that the morpheme encoding sensory or visual evidentiality in Lhasa Tibetan also marks mirativity is not found fully convincing because it is not placed within the context of a full grammar of the language. According to Aikhenvald (2004: 210), any evidential except for visual and firsthand, can acquire mirative extensions. These extensions are a frequent feature of the non-firsthand form in small systems, and of inferred in larger systems.
[…] The emerging correlation is intuitively right: new and unexpected knowledge appears to be interconnected with something one makes inferences about but has not witnessed or hardly has under control.
The Moldavian Hungarian facts suggest a differenttrivialcorrelation, which holds for all languages with a two-term evidential system contrasting an evidentially marked and an evidentially neutral form. Namely, if the evidential system can also express mirativity, it is the evidentially marked form that has a mirative use.
In Moldavian Hungarian, the mirative use of the evidential -a/e appears to be elicited in the case of directly experienced events that are not only unexpected but also trigger an emotional reactionmostly sorrow, anger, or exasperationin the speaker. An emotional component has been pointed out in the licencing of mirative morphemes in some other languages, as well. As observed by DeLancey (2012: 558), unexpected information often provokes some kind of emotional reaction, and it is inevitable that some emotional value will sometimes attach to a mirative construction. DeLancey points to a number of descriptions that refer to implications of disapproval or other negative attitudes associated with mirative constructions, e.g., Hein (2007) and So-Hartmann (2009: 293-294). In Tsakhur (Tatevosov and Maisak 1999: 232-233), the non-firsthand evidential may acquire a mirative connotation if something happens contrary to the speaker's expectation and much to their regret. Sapir (1922: 200), too, mentions anger among the conditions licencing the mirative use of the inferential in Takelma: it is frequently used "in rhetorical questions of anger, surprise, wonder, and discovery of fact after ignorance of it for some time". Mirative exclamations in Moldavian Hungarian often function as self-addressed rhetorical questionssimilarly to rhetorical questions marked by mirative forms e.g. in !Xun (König 2013), Magar (Grunow-Hårsta 2007), and Cantonese (Matthews 1998: 330-331), languages reported by Aikhenvald (2004: Ch. 6, 2012.
As I will show in the following section, the atypical properties of evidentiality in Moldavian Hungarian are a result of its historical evolution.
3 The origin of the Moldavian Hungarian evidential system V+t/tt+AGR AUX+a mond-t-am val-a say-PRF-1SG be-PST ʻI had saidʼ The suffix -a/e, descendant of the Uralic past-tense suffix, functioned as the marker of the general, narrative past, the past tense of events not linked to the present (hence it is glossed as PST in the Old Hungarian examples). In the late Old Hungarian period, however, it faced a competitor, the suffix -t. The suffix -t originally encoded the perfect aspect (hence it is glossed as PRF in the Old Hungarian examples). Unless followed by an -a/e-marked temporal auxiliary, t-marked verbs denoted past events with present relevance (Abaffy 1992;Szarvas 1872). Notice the different functions of the narrative past (used by the story teller giving account of a past event) and the present perfect (used by a participant of the same situation facing the present result of the past event) in the following example: (18) Orpha megapol-a o̗ nap-a-t & mėǵfordol-a. Rvt Orpah kiss-PST. 3SG she mother.in.law-3SG-ACC and turn-PST.3SG Ruth ėggèso̗ l-è o ̗ nap-a-ual Ki-nc ̣ mōd-a Noemi unite-PST.3SG she mother.in.law-3SG-with who-DAT tell-PST.3SG Noemi Ime te rokon-od mėǵfordol-t. lo you relative-2SG turn-PRF.3SG ʻOrpah kissed her mother-in-law and turned round. Ruth held onto her mother-in-law. Noemi told her: Lo, your relative has turned round.' (Bécsi Codex: 2) V+AGR combined with an -a-marked temporal auxiliary represented past imperfect, i.e., continuous or habitual past:

(19)
Es ez bezed-ek-en … zent lelek bezzell-e ysteny edes and this speech-PL-SUP holy spirit speak-PST.3SG divine sweet bezed-ek-ett … hogy mend czudal-yak-ual-a Es halgat-yak-ual-a speech-PL-ACC that all admire-3PL-be-PST and listen-3PL-be-PST ew-tett he-ACC ʻAnd in these speeches [of Saint Francis] the holy spirit spoke divine sweet speeches … so that they were all admiring him and listening to him' (Jókai Codex: 37) A perfect, -t-marked verb followed by an -a-marked temporal auxiliary expressed past perfect tense: (20)És megemlékez-é-k Péter az igé-ről, ki-t mondo-tt and commemorate-PST-3SG Peter the word-about which-ACC say-PRF.3SG val-a be-PST 'And Peter commemorated the word that he had said.' (Müncheni Codex: 103) The tense system outlined above started disintegrating in the late Old Hungarian period already. It was still more or less complete in documents recording 'bequethed' texts inherited from earlier times, i.e., in poems, and in the prose of codices. (Of the 46 Hungarian-language codices from the 14th, 15th and early 16th centuries that survived the Turkish occupation of Hungary in 1526, all but one are copies of older texts, mostly Bible translations, monastic rules, prayers, and legends of saints.) In spontaneously created texts, mostly letters, however, the -a/e suffix was already losing ground to -t, which was developing from the marker of past events with present relevance into a marker of past events. Compare the distribution of past tenses in three different text types from the 14th, 15th and early 16th centuries (Abaffy 1992: 163): In poems and codices, mostly representing copies of older exemplars, we find twice as many occurrences of -a/e-marked verbs than of -t-marked verbs. In letters representing spontaneous language use, the proportion of these tenses is reversed: there are more than twice as many -t-marked verbs as -a/e-marked verbs. The change in proportions went hand-in-hand with a functional change: -t gradually took over the role of the general past-tense suffix. The development of perfect into general past, i.e., the simplification of the [+past, +present relevance] feature complex to [+past] is not unique to Hungarian; it also happened e.g. in the case of the French passé composé (Caudal 2015).
As a consequence of this process, the -a/e suffix, the former narrative/general past marker, was gradually constrained to particular functional niches in Middle Hungarian. Abaffy (1992) already observed a new function of -a/e in a few Late Old Hungarian documents; she noticed that -a/e could also denote recent past in reported direct speech. The following Old Hungarian example illustrates both the regular and the newly emerging function of -a/e: the suffix marks narrative past on the reporting verb, and recent past on the reported verb.
(21) es mōd-a … "Ez az ki-ro̗ l mond-e-cèn vtann-am io̗ and say-PST.3SG this that who-about say-PST-1SG I after-1SG come.3SG felʼlʼ-o̗ l" above-from ʻAnd said … "This is the one about whom I have said that he will come from above after me"ʼ (Müncheni Codex: 85)  Abaffy (1992: 153, 166) claims that the marking of recent past by -a/e emerged as a dialectal phenomenon in codices with ties to Eastern Hungary.

3.2
The -a/e suffix in Middle Hungarian (1526-1772): decreasing frequency, pragmatic conditions on its use Whereas Old Hungarian marked both tense and viewpoint aspect, i.e., +/-perfectivity, the emergence of a system of situation aspect marking by means of telicizing verbal particles led to the disappearance of viewpoint aspect marking in the Middle Hungarian period (É. Kiss 2016). The language was slowly returning to the Proto-Ugric and Proto-Uralic system of two simple tenses: past and non-past, with past tense marked by the suffix -t instead of the original -a/e. Compare the distribution of the four past tenses in Old Hungarian, shown in Table 1, with their distribution in a corpus of 1,225 letters from the period 1526-1700 (Mohay 2018: 71), shown in Table 2.
The commonness of the -t-past and the infrequence of the -a/e past is also confirmed by Halász's (2021) data, who examined the percentages of the -t-marked (vol-t) and -a-marked (val-a) past-tense forms of the copula used as a discrete (main) verb among the tokens of the 1,050,000 word Old and Middle Hungarian corpus of informal language use (http://tmk.nytud.hu). Compare their changing proportions in the 16th, 17th and 18th centuries, shown in Figure 1:  Parallel with -t taking over the role of general past-tense suffix, the distribution of -a/e gradually came to be restricted to specific contexts. Owing to its relative rarity, the -a/e past assumed an elevated stylistic value and tended to be used to highlight important, distinguished events. This tendency has been observed by several authors analysing the Middle Hungarian tense system. Németh (2001) compared the distribution of past tenses in two late 16th-century corpora of similar sizes: private letters and a memoir appended to an entreaty for a land grant, emphasizing the authorʼs services of 65 years for an aristocratic family. She found that the -a/e past, making up 59.4% of all past tenses in the memoir and only 7.1% in the private letters, was used in the case of significant, often majestic, happenings. Zsilinszky (2005), analysing 17thcentury manor court records, observed that the -a/e past was often used for dramatic purposes, highlighting incidents that represented a turn in a series of events. Her observations recall the characterization of mirative evidentials in some present-day languages. For example, the mirative marker in !Xun can mark the main point of the story (König 2013). In Cantonese, the mirative can be used to express notable information (Matthews 1998: 330-331).
The pragmatic contexts of the -a/e past have been investigated most thoroughly by Mohay (2018Mohay ( , 2020Mohay ( , 2021. In addition to 1,225 Middle Hungarian letters, she studied witness testimonies of witch trials, relying on the Old and Middle Hungarian corpus of informal language use (https://tmk.nytud.hu/). The Middle Hungarian letters analysed by her show a function of the -a/e past that was already present in some Old Hungarian codices as a dialectal phenomenon (Abaffy 1992: 163): the marking of recent past, i.e., the period within a few days before the speech time (Mohay 2018). The typical time adverbials used with -a/e-marked verbs include tegnap 'yesterday', e napokban 'these days', ez órában 'this hour', most 'now', as illustrated in (22a, b). (Since in Middle Hungarian, -t is already a general past-tense marker, whereas -a/e has a varying role, I gloss them as PST and APST, respectively.) (22) a. ez ora-ban erkez-e-k leuel-em this hour-in arrive-APST-3SG letter-1SG 'My letter has arrived at this hour.' (1621, Mohay 2018: 79) b. En már a dolog-rol beszéll-ett-em vol-t, most uiyobban I already the matter-about speak-PST-1SG be-PST now again beszéll-é-k speak-APST-1SG 'I had already spoken about the matter, and now I have spoken about it again.' (1668( , Mohay 2018 In the Middle Hungarian period investigated by Mohay (2018), the recent past marking role of -a/e is not a dialectal phenomenon anymore; she claims that it is attested in letters from all over Hungary.
The letters also contain contexts where the -a/e past encodes significant unexpected events, i.e., it has a kind of mirative role: (23) Ez ell mult ennehani esztendó-ben, igen el this PRT past few year-in quite off ueu-e az ur Isten haza-nk-nak gonduiseló-y-tt, take-APST.3SG the lord God country-1PL-DAT guardian-POSSPL-ACC nemzet-únk-nek io gonduiseló-i-t kiszed-e koszull-únk nation-1PL-DAT good guardian-POSSPL-ACC remove-APST.3SG from-1PL 'In these past few years, the Lord completely took away our countryʼs guardians, it removed our nationʼs good guardians from among us.' (1609( , Mohay 2018 In the witch trial testimonies analysed by Mohay (2021), the most typical function of the -a/e past is the encoding of significant events that affected the speaker personally. 21.24% of the -a/e-marked verbs in court records belong to this type. Mohay set up the following five categories of personally experienced significant events typically denoted by -a/e-marked verbs: (i) curse, disease, symptoms of a disease, abuse, death, e.g.: (24) az fi-a-t megh uer-t-em uol-t az megh fog-ot the son-3SG-ACC PRT beat-PST-1SG be-PST the PRT catch-PSTPTCP Azoni-nak, az giermek megh mond-ott-a otthon az an-ia-nak woman-DAT the child PRT say-PST-3SG home the mother-3SG-DAT es ugi jar-e-k harmad nap muluan, hogi mind az and so come.off-APST-1SG three day after that both the ket lab-om özue sugorod-e-k. two leg-1SG together shrink-APST-3SG 'I had beaten the son of the captured woman, the child told it to her mother at home, and it happened to me after three days that both of my legs shrank.' (1629( , Mohay 2021 (ii) other kinds of negative, threatening events, damage: könyök-ön feküd-t-em, az pipa is szá-m-ban vól-t, nagj elbow-on lie-PST-1SG the pipe too mouth-1SG-in be-PST.3SG very hirtelen az Falu felől nagj zugas es-é-k suddenly the village from great buzzing happen-APST-3SG 'I was lying on my elbow, also the pipe was in my mouth, all of a sudden there was a strong buzzing from the village' (1723, Mohay 2021: 100) (iii) recovery, relief of bodily symptoms: (26) más nap mingyárt tulaidon ere-i-vel föll kel-e, next day right.away own strength-POSSPL-with up get-APST.3SǴ es ki mén-e, holott an-nak előtt-e Lábai-ra and out go-APST.3SG though that-DAT before-3SG feet.3SG-on sem áll-hat-ott not.even stand-POSSIB-PST.3SG 'and so the next day he got up right away by his own strength and went out, even though before that he could not even stand on his feet' (1743, Mohay 2021: 101) (iv) immediate precursors of a turn of events: tud-od, hogy kér-t-em harom Mariás-t tül-ed kutya leány-a, know-2SG that ask-PST-1SG three marias-ACC from-2SG dog daughter-3SǴ es nem ad-á-l, lát-od-é mint jár-á-l and not give-APST-2SG see-2SG-Q how come.off-APST-2SG 'You know that I asked you, dogʼs daughter, for three Maria-coins, but you did not give any; do you see what happened to you?' (1729( , Mohay 2021 (v) emotional reactions to a turn of events: (28) azzal el keseredvén a tornácz-bon ki fordul-t,és then PRT growing.desperate the porch-in out turn-PST.3SG and köservessen sir-a bitterly cry-APST.3SG 'Then having grown desperate, he turned in the porch and cried bitterly.' (1748, Mohay 2021: 102) Mohay (2021: 88) concluded that the most prominent role of the -a/e past in witch trial testimonies is the expression of personal affectedness and experience, i.e., the expression of a kind of direct evidentiality.
In sum: statistical data indicate that the -a/e-marked past lost ground to the -t-marked past in Middle Hungarian. The use of -a/e was gradually restricted to particular contexts, i.e., to verbs describing recent past events and to verbs describing significant events that were unexpected and/or affected the speaker personally. Its frequent use in contexts describing significant unexpected events provides evidence of its mirative value, whereas its growing use in contexts describing events that affected the speaker personally indicates its emerging evidential role. At the same time, the mirative and evidential uses of -a/e are merely tendencies, which are only present in a subset of all occurrences.

The -a/e-past in Modern Hungarian (1772-): areal reduction; the grammaticalization of the conditions of its use
The functional reduction of the -a/e past also continued in the Modern Hungarian period and was accompanied by areal reduction. The suffix gradually disappeared from north to south and from west to east, and eventually it only survived in the eastern dialects as the marker of witnessed immediate past and witnessed mirative past.
Interestingly, 18th-, 19th-and early 20th-century Hungarian grammars, influenced by grammars of Latin, only recognize the temporal function of -a/e. They describe it as the marker of immediate past, calling it the suffix of alig múlt ʻbarely pastʼ or fél múlt ʻhalf pastʼ (e.g., Magyar grammatika 1795 ;Schedel 1846;Szarvas 1872;Simonyi 1895;Klemm 1928). The poet János Arany, however, also noticed its mirative role (Arany 1879). He described -a/e as a suffix used in eastern and central Hungary to encode unexpected recent events. The recency condition of its use can be neutralized, however, if unexpectedness is accompanied by anger, sorrow, or exasperation. Arany cited the following example of -a/e expressing mirativity without recency: (29) (Two friends meet after a long time, and share the most important news since their last encounter.) "Meg-hal-a szegény Péter is!" "Meg biza! Tavaly hal-t PRT-die-APST.3SG poor Peter too PRT indeed last.year die-PST.3SG meg." PRT "Poor Peter died, too!" "Yes, he did. He died last year." When Peterʼs death is mentioned for the first time, the -a/e-marked verb describes it as an unexpected exasperating piece of news. In the second clause, his death is a presupposed fact, and the specification of the time of his death is neutral information, with the verb bearing the general past-tense suffix.
The evidential role of the Hungarian -a/e suffix was first noticed by a Turkologist, Gábor Bálint. His Kazan Tatar grammar (1877) mentionsby way of analogythat Secler Hungarians (the western neighbours of Moldavian Hungarians) use the -a/e past to express actions that they carried out or witnessed themselves. However, Bálintʼs observation remained unnoticed by Hungarian grammarians.
Hungarian grammars from the middle of the 20th century (e.g., Tompa 1962) do not discuss the -a/e past among the actively used past-tense suffixes any more. Vámszerʼs (1972) dialectal study still reported the sporadic use of -a/e as a suffix of recent past in Kalotaszeg, an Eastern region of the Hungarian language area and it also described the condition of direct experience in licencing its use. More recent dialectal studies, e.g. Kiss (2003), mention the survival of the -a/e past in the Eastern Hungarian Secler and Moldavian dialects without discussing its special function.
The notions of direct experience and witnessing, associated with -a/e by Bálint (1877), emerged again in Tolcsvai Nagyʼs (2017) cognitive-functional analysis of the use of past tenses in a letter written by János Arany in 1847, and in the above cited studies of Mohay (2018Mohay ( , 2020Mohay ( , 2021.
4 The evolution of the evidential-mirative function of -a/e In Old Hungarian, the -a/e suffix marked past events unrelated either to the present or to the speaker. In present-day Moldavian Hungarian, by contrast, it marks events that happened right before the moment of speech, and which either involved the speaker or were witnessed by him or her. The mirative extension of narrative past has also been reported from Hindi/Urdu (Montaut 2006), but the developmental path from evidentially neutral narrative past to immediate past encoding mirativity and direct evidentiality is, to the best of my knowledge, unique, and hence in need of explanation.
The functional reinterpretation of the -a/e past must have been initiated by its relative scarcity in the Middle Hungarian period and its losing ground to the -t-marked past. The infrequency of the -a/e past gave a/e-marked verbs an elevated stylistic value as a result of which they came to be used to highlight significant, emotionally involving eventsas observed by Németh (2001), Zsilinszky (2005), and Mohay (2018Mohay ( , 2021.
It seems plausible that its rarity and the resultant connotation of exception resulted in the evolution of a mirative overtone of the -a/e past, expressing that the event denoted by the -a/e-marked verb was unexpected and had a significant impact upon the speaker. An unexpected event is particularly likely to make a significant impact upon the speaker if the speaker is a participant, or at least a witness, of itwhich may have extended the function of -a/e to direct evidentiality. Unexpected, emotionally engaging impacts usually involve recency, which explains why the 'past' meaning of -a/e came to be restricted to 'recent past' or even 'immediate past'. These considerations suggest the following hypothetical functional evolution of -a/e: (rare) past > emotionally loaded past > mirative past > witnessed mirative past > witnessed recent past In the case of the a-marked imperfect auxiliary, the evolution stopped before reaching its final stage, hence the verb form 'V+Agr val+a' expresses +/-recent witnessed imperfect past. As for the timeline of these changes, the marking of recent past emerged as a secondary function of -a/e in documents related to Eastern Hungary in the late Old Hungarian period. This means that the semantic development outlined in (30) must have taken place at least in Eastern Hungary by the 16th century. The Middle Hungarian data show the gradual fading of the general past marking role of -a/e and the areal spreading of its new 'mirative past', 'witnessed mirative past' and 'witnessed recent past' marking functions. In the Middle Hungarian period mirativity, direct evidentiality and recency were still pragmatic aspects of the use of the -a/e past; they were not evoked in all contexts. These functions grammaticalized as inherent parts of the meaning of -a/e in Early Modern Hungarian. At present, only the last two stages of the evolution in (30) survive, and they are restricted to the easternmost dialects of the language.

Conclusion
In the Moldavian dialect of Hungarian, the complex tense system of Old and Middle Hungarian, encoding tense and aspect, has developed into a tense system also encoding evidentiality and mirativity. Moldavian Hungarian evidentiality is atypical in at least three respects. First, it is a two-term evidentiality system based on the opposition of direct evidentiality and the lack of evidentialitywhereas other twoterm evidential systems involve firsthand versus non-firsthand (A1 type), nonfirsthand versus everything else (A2), reported versus everything else (A3), sensory evidence versus reported (A4), and auditory versus everything else (A5) oppositions (Aikhenvald 2004). Second, mirative meaning has the same exponent as direct evidentialitywhereas in small evidential systems usually non-firsthand acquires a mirative overtone (Aikhenvald 2012: 465). Third, the mirative function of the -a/e past appears to have emerged before its evidential interpretation, whereas it is evidentials that generally develop mirative extensions and not the other way round (Aikhenvald 2012: 465).
The atypical properties of Moldavian evidentiality must be consequences of its historical evolution. The first and second of these properties follow from the third one, the primacy of the mirative function of the -a/e past. As for the second property, mirativity can, in principle, be associated with new information independently of whether the information has been obtained by direct witnessing or inference. Therefore, the mirative -a/e could have assumed a meaning component of either direct or inferential evidentiality. However, the -a/e-past-tense suffix was a competitor of -t, originally the marker of the perfect. It was the -t suffix, encoding past events with a present result or present relevance, that had a latent inferential component, a potential source of indirect evidentiality. Hence the option open for the evidential extension of the mirative -a/e past was direct evidentiality.
The unusual two-member evidentiality system based on the opposition of direct evidentiality and the lack of evidentiality must be a consequence of the same facts. That is, when the mirative and evidential functions of the -a/e past emerged, the perfect, which is the most likely source of inferential evidentiality, was still associated with the suffix -t, which started developing into a general past-tense marker. The loss of the general past-tense marking role of -a/e helped it assume new mirative and evidential functions. In the case of -t, the opposite happened: the generalization of -t into a general past-tense marker neutralized the feature that could have developed into inferential evidentiality.