PrevHist: A dataset of Old and Middle Hungarian preverb constructions

Overview

PrevHist is a dataset of preverb-verb constructions from Old and Middle Hungarian, covering a time span of approximately nine centuries (896–1772). It contains 68,458 records extracted automatically from two historical corpora and is annotated for a wide range of morphosyntactic features and metadata. It is designed to be compatible with PrevDistro, a large-scale dataset of preverb-verb constructions in Present-day Hungarian (Kalivoda, 2022). Both resources share identical design principles and column names wherever possible, making aggregation straightforward for researchers wishing to extend their analysis toward the modern end of the diachronic spectrum.

Source Corpora

Records were extracted from two corpora:

OHC – Old Hungarian Corpus (Simon, 2014). Designed as a comprehensive sample of all surviving Old Hungarian sources (896–1526) plus some Middle Hungarian Bible translations. Texts are predominantly religious literature. The dataset was built from the morphosyntactically annotated subset of OHC, which amounts to approximately 300,000 tokens. This yields 12,781 records in PrevHist.

MHC – Old and Middle Hungarian Corpus of Informal Language (Novák, Gugán, Varga, & Dömötör, 2018). Focused on the vernacular of the Middle Hungarian period (1526–1772), primarily court records of witch trials and private letters. The whole corpus (~1.5 million tokens) is normalized and morphosyntactically annotated, and was used in its entirety. This yields 55,677 records in PrevHist.

Dataset

File
Format
Encoding
Line endings
Records

PrevHist.tsv
Tab-separated values (TSV)
UTF-8
Unix/Linux (LF)
68,458

The first row is the header. Missing or not applicable values are marked with an underscore (_).

#
Column
Description

1
sid
Record ID (integer).

2
constype
Construction type. FIN = finite verb construction; all other values are verbal derivational suffixes (e.g., -ni = infinitive, -vA = adverbial participle, -Ás = nominalization, etc.). Capital letters in suffix codes stand for archiphonemes (cover symbols for vowel harmony variants).

3
subtype
Construction subtype. Encodes the combination of construction type and word order: prefixed_* = preverb is morphologically fused with the verb (i.e., the preverb appears immediately before the verb stem); *_inverted = preverb follows the verb; *_discontinuous = preverb and verb are separated by one or more intervening elements.

4
prevpos
Preverb position relative to the verb. 0 = prefixed (immediately preceding, fused); positive integers = number of tokens after the verb; negative integers = number of tokens before the verb (other than the immediately preceding position).

5
prev
Preverb lemma (e.g., meg, el, ki, fel, be, le).

6
verb
Verb lemma.

7
intervening
Lemmas of words intervening between the preverb and the verb stem, space-separated. _ if none.

8
incorporation
The incorporated noun, if present (verbal incorporation construction); _ otherwise.

9
actform
The actual form of the preverb-verb construction as it appears in the normalized text.

10
orig_left
Left context in original (non-normalized) orthography.

11
orig_kwic
Keyword in context (the preverb-verb construction) in original orthography.

12
orig_right
Right context in original orthography.

13
norm_left
Left context in normalized orthography.

14
norm_kwic
Keyword in context in normalized orthography.

15
norm_right
Right context in normalized orthography.

16
corpus
Source corpus: OHC or MHC.

17
docid
Document ID within the source corpus.

18
year
Year of writing of the source document. _ if unknown.

19
profile
Register/text type. Values: religious-formal, witch_trial, letter-informal, letter-formal, letter-formal|letter-informal.

20
hunmeta
Additional metadata from MHC in Hungarian, including document reference, date, sender, recipient, and social variables. _ for OHC records.

Limitations

The two source corpora differ substantially in size, register, and source diversity. OHC is relatively small and composed almost exclusively of formal religious texts. MHC is much larger and represents informal vernacular language across numerous documents and several decades. This asymmetry makes direct cross-period comparison non-trivial. Researchers should exercise caution when interpreting diachronic trends, as differences may reflect corpus composition rather than genuine language change. These limitations are inherent to the current state of Hungarian historical corpora rather than to the dataset itself.

Funding Statement

This project has been supported by the OTKA PD project No. 142317 funded by the Ministry of Culture and Innovation of Hungary from the National Research, Development and Innovation Fund, financed under the PD 22 funding scheme.

References

Kalivoda, Á. (2022). PrevDistro: An open-access dataset of Hungarian preverb constructions. Acta Linguistica Academica, 69(4), 549–563. https://doi.org/10.1556/2062.2022.00578

Novák, A., Gugán, K., Varga, M., & Dömötör, A. (2018). Creation of an annotated corpus of Old and Middle Hungarian court records and private correspondence. Language Resources and Evaluation, 52(1), 1–28. https://doi.org/10.1007/s10579-017-9393-8

Simon, E. (2014). Corpus Building from Old Hungarian Codices. In K. É. Kiss (Ed.), The evolution of functional left peripheries in Hungarian syntax (pp. 224–236). Oxford University Press.