AthDGC: An Open Diachronic Greek Treebank with Indo-European Parallels

Anastasia Tsiropina; Dag Haug; Eleni Plakoutsi; Evangelos Argyropoulos; Kiki Nikiforidou; Leonid Kulikov; Nikolaos Lavidas; Sofia Chionidi; Theodoros Michalareas; Vassileios Symeonidis

AthDGC: An Open Diachronic Greek Treebank with Indo-European Parallels

T0 review · 2 major / 0 minor · reviewed 2026-07-03 · grok-4.3

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read AthDGC supplies the first openly licensed dependency treebank of Greek across eight diachronic periods under one PROIEL schema with parallel alignments to four other ancient languages.

desk verdict AthDGC is a first-of-its-kind open multi-period Greek treebank with alignments, but the uniform parser lacks any accuracy validation across periods. read the letter →

arxiv 2606.15510 v2 pith:3AJG7RW3 submitted 2026-06-13 cs.CL cs.DL

Nikolaos Lavidas , Kiki Nikiforidou , Dag Haug , Leonid Kulikov , Vassiliki Geka , Vassileios Symeonidis , Theodoros Michalareas , Sofia Chionidi

show 3 more authors

Anastasia Tsiropina Eleni Plakoutsi Evangelos Argyropoulos

This is my paper

classification cs.CLcs.DL

keywords GreektreebankdiachroniccorpusdependencyparsingPROIELschemaNewTestamentalignmenthistoricallinguisticsIndo-Europeanparallelsopendataset

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper presents AthDGC as an open workflow and dataset that annotates Greek texts from Archaic through Modern periods with dependency parses. It places all eight stages in a single PROIEL XML schema and adds verse-level alignments of the New Testament to Latin, Gothic, Old Church Slavonic, and Classical Armenian. The resource builds directly on the existing PROIEL family and uses Stanford Stanza for parsing plus LaBSE and AwesomeAlign for alignments. A reader would value it because prior resources covered only isolated periods or lacked consistent licensing and cross-lingual links, limiting large-scale diachronic and comparative work.

What carries the argument

The PROIEL XML 2.0 schema, which supplies a uniform dependency annotation format across all periods and languages, together with the Stanford Stanza PROIEL-trained parser and the LaBSE/AwesomeAlign alignment pipeline.

What would settle it

A manual audit of several hundred randomly sampled sentences from each of the eight periods that finds dependency-label error rates above 15 percent would show the workflow does not meet the consistency requirement.

Watch

Extended reading notes

Core claim

AthDGC is an open, end-to-end workflow and dataset that constitutes the first openly licensed dependency-parsed treebank of Greek spanning eight diachronic periods (Archaic, Classical, Koine, Late Antique, Byzantine, Late Byzantine, Early Modern, and Modern Greek) under a single PROIEL XML 2.0 schema, with verse-level cross-alignment of the New Testament to Latin (Vulgate), Gothic (Wulfila), Old Church Slavonic (Marianus), and Classical Armenian.

Load-bearing premise

The automated parsing and alignment procedures produce sufficiently accurate and consistent annotations across all eight periods without major period-specific errors.

Editorial extensions

If this is right

Researchers can now run uniform syntactic queries on Greek texts from the eighth century BCE to the present day.
Verse alignments enable direct comparison of syntactic structures in the New Testament across five Indo-European languages.
The single schema removes the need to reconcile incompatible annotation formats when studying language change.
Open licensing permits reuse for training historical language models or for further manual correction.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The same workflow could be applied to additional witnesses or languages if the underlying models generalize beyond the current training data.
Quantitative studies of syntactic drift in Greek could now be performed on a scale that isolated-period treebanks did not allow.
The resource supplies a test bed for checking whether cross-lingual embedding methods remain reliable when applied to historical language stages.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

read the letter

The paper's core contribution is releasing the first openly licensed dependency treebank for Greek that covers eight periods from Archaic to Modern under a single PROIEL schema, plus verse alignments for the New Testament to Latin, Gothic, Old Church Slavonic, and Armenian. It extends the existing PROIEL family with a workflow using Stanza for parsing and LaBSE/AwesomeAlign for alignments, and it ships samples plus the toolkit now, with full data after audit.

That scope and licensing choice is genuinely new and fills a gap for diachronic work. The schema consistency and cross-lingual links are practical additions that researchers in historical linguistics or digital humanities could use directly.

The main weakness is the annotation quality. A single Stanza model trained on Koine is applied to all periods, yet Archaic and Modern Greek differ sharply in word order, case, and subordination. The abstract and release notes give scale and schema details but no UAS/LAS figures, per-period error analysis, or linguist-checked samples. The v0.5 audit is still pending, so the claim that this is a usable high-quality resource rests on an untested assumption.

This is a data paper, not a methods or theory one. It is worth bringing to a reading group for anyone who works with Greek treebanks or alignments, since the resource itself could be valuable once validated. I would not cite the paper itself unless I were using the data, but the data release could be citable later.

It deserves peer review as a resource contribution, provided the authors add the missing accuracy metrics and validation details. Without those, the usability case stays thin.

Referee Report

2 major / 0 minor

Summary. The paper presents AthDGC (Athens-PROIEL), an open workflow and dataset that produces the first openly licensed dependency-parsed Greek treebank spanning eight diachronic periods (Archaic, Classical, Koine, Late Antique, Byzantine, Late Byzantine, Early Modern, Modern) under a single PROIEL XML 2.0 schema, together with verse-level cross-alignments of the New Testament to Latin, Gothic, Old Church Slavonic, and Classical Armenian. Annotation is performed with a single Stanford Stanza model trained on the PROIEL Koine reference; sentence alignment uses LaBSE and word alignment uses AwesomeAlign on multilingual BERT. The v0.4 release supplies curated samples and the toolkit; the full corpus is under v0.5 audit, with scale statistics to appear in the release notes.

Significance. If the produced annotations and alignments are shown to be sufficiently accurate and consistent, the resource would constitute a valuable addition to diachronic and comparative Indo-European syntax research by providing the first openly licensed multi-period Greek treebank with parallel alignments. The open-source workflow and adherence to the established PROIEL schema are clear strengths.

major comments (2)

[Abstract / annotation workflow] Abstract and annotation-workflow description: no UAS/LAS figures, error analysis, or per-period validation results are reported for the single PROIEL-trained Stanza model applied uniformly across all eight periods. Archaic freer word order and Modern analytic constructions are known to differ from the Koine training data, so the absence of any accuracy metric leaves the claim of a usable high-quality resource without empirical support.
[Abstract / alignment description] Alignment procedures (LaBSE sentence-level and AwesomeAlign word-level): no quantitative evaluation, manual audit samples, or period-specific reliability checks are provided, despite the claim that these alignments are suitable for downstream research.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments highlighting the need for empirical validation. We address each major comment below and commit to revisions that strengthen the manuscript without misrepresenting the current v0.4 release status.

read point-by-point responses

Referee: [Abstract / annotation workflow] Abstract and annotation-workflow description: no UAS/LAS figures, error analysis, or per-period validation results are reported for the single PROIEL-trained Stanza model applied uniformly across all eight periods. Archaic freer word order and Modern analytic constructions are known to differ from the Koine training data, so the absence of any accuracy metric leaves the claim of a usable high-quality resource without empirical support.

Authors: We agree that the manuscript lacks the requested quantitative metrics and error analysis. The current version focuses on the open workflow and v0.4 curated samples, with full annotation under v0.5 audit. In revision we will add a dedicated validation subsection reporting UAS/LAS on the PROIEL Koine held-out set, plus manual spot-check error analysis on samples from Archaic and Modern periods, while explicitly noting the limitations arising from diachronic syntactic differences and that comprehensive per-period figures will appear in the v0.5 release notes. revision: yes
Referee: [Abstract / alignment description] Alignment procedures (LaBSE sentence-level and AwesomeAlign word-level): no quantitative evaluation, manual audit samples, or period-specific reliability checks are provided, despite the claim that these alignments are suitable for downstream research.

Authors: We acknowledge that no quantitative alignment evaluation or period-specific checks are currently reported. The alignments rely on established models (LaBSE for sentences, AwesomeAlign on mBERT for words), but the manuscript does not include supporting metrics. In the revised version we will insert a short evaluation subsection with alignment precision/recall estimates derived from manual inspection of a random sample of New Testament verses, together with a clear statement that the alignments are offered as an initial resource and that downstream users should perform their own validation. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; workflow relies on external tools and prior schema

full rationale

The paper presents a data-creation workflow and dataset release that applies the existing Stanford Stanza PROIEL-trained parser, LaBSE sentence embeddings, and AwesomeAlign word alignment to Greek texts across periods, extending the independently established PROIEL XML schema from Haug and Johndal 2008 and Eckhoff et al. 2018. No equations, fitted parameters, predictions, or derivations appear; the central claims concern scale, schema compliance, and open licensing rather than any result that reduces by construction to its own inputs. Self-citations to the PROIEL family are load-bearing only for schema reuse, not for any uniqueness theorem or ansatz that would create circularity. The v0.5 audit mention and lack of per-period UAS/LAS figures raise questions of empirical validation but do not constitute circular reasoning.

Assumptions & free parameters 0 free parameters · 1 assumptions · 0 invented entities

The paper introduces no free parameters, invented entities, or ad-hoc axioms; it applies the existing PROIEL schema and off-the-shelf tools (Stanza, LaBSE, AwesomeAlign) to new data.

assumptions (1)

domain assumption The PROIEL XML 2.0 schema remains suitable for consistent dependency annotation of Greek across all eight listed periods.
Invoked by the decision to use a single schema for the entire diachronic range without period-specific adjustments mentioned.

how reviews work

0 comments

Cite this review

Pith. "Pith review of AthDGC: An Open Diachronic Greek Treebank with Indo-European Parallels." pith.science (2026). https://pith.science/paper/3AJG7RW3

@misc{pith2026260615510,
  author       = {Pith},
  title        = {Pith review of: AthDGC: An Open Diachronic Greek Treebank with Indo-European Parallels},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3AJG7RW3}},
  note         = {Machine review of arXiv:2606.15510}
}

read the original abstract

AthDGC ("Athens-PROIEL") is an open, end-to-end workflow and dataset. It is, to the best of our knowledge, the first openly licensed dependency-parsed treebank of Greek that spans eight diachronic periods, namely Archaic, Classical, Koine, Late Antique, Byzantine, Late Byzantine, Early Modern, and Modern Greek, under a single PROIEL XML 2.0 schema, with verse-level cross-alignment of the New Testament to Latin (Vulgate), Gothic (Wulfila), Old Church Slavonic (Marianus), and Classical Armenian. AthDGC builds on the PROIEL Treebank Family (Haug and Johndal 2008; Eckhoff et al. 2018), which established the schema and the Koine-Greek reference set for the project. Annotation uses the Stanford Stanza PROIEL-trained workflow; sentence-level alignment uses LaBSE, a multilingual sentence-embedding model; word-level alignment uses multilingual-BERT attention through the AwesomeAlign procedure. The v0.4 release provides curated samples and the open-source toolkit; the full annotated corpus partitions remain under v0.5 audit on the Greek national HPC. Quantitative scale, per-witness verse counts, and per-period annotated-row counts are reported in the v0.5 release notes, after the audit pass completes. Concept DOI: 10.5281/zenodo.20439182.

Figures

Figures reproduced from arXiv: 2606.15510 by the authors.

**Figure 1.** A single PROIEL XML 2.0 sentence record as emitted by the AthDGC autopass for Iliad 1.1, extracted from corpus/merged/language=grc/period=archaic/Homer__Iliad.jsonl on the v0.4 release. Five tokens; one tree; six annotation layers per token (form, lemma, pos, morph, head+relation, info_status+antecedent_id); the same record format runs from Homer through Modern Greek and across the four Indo-European parallels alig… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

11 extracted references · 11 canonical work pages

[1]

Bary, C., & Haug, D. T. T. (2011). Temporal anaphora across and inside sentences: The function of participles. Semantics and Pragmatics, 4(8), 1–56. https://doi.org/10.3765/sp.4.8

work page doi:10.3765/sp.4.8 2011
[2]

Dou, Z.-Y., & Neubig, G. (2021). Word alignment by fine-tuning embeddings on parallel cor- pora. In Proceedings of the 16th Conference of the European Chapter of the Association for Com- putational Linguistics: Main Volume (pp. 2112–2128). Association for Computational Linguistics. https://doi.org/10.18653/v1/2021.eacl-main.181

work page doi:10.18653/v1/2021.eacl-main.181 2021
[3]

M., Bech, K., Bouma, G., Eide, K., Haug, D

Eckhoff, H. M., Bech, K., Bouma, G., Eide, K., Haug, D. T. T., Haugen, O. E., & Jøhndal, M. (2018). The PROIEL treebank family: A standard for early attestations of Indo-European languages. Language Resources and Evaluation, 52(1), 29–65. https://doi.org/10.1007/s10579-017-9388-5

work page doi:10.1007/s10579-017-9388-5 2018
[4]

Feng, F., Yang, Y., Cer, D., Arivazhagan, N., & Wang, W. (2022). Language-agnostic BERT sen- tence embedding. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (pp. 878–891). Association for Computational Linguistics. https://doi.org/10.18653/v1/2022.acl-long.62

work page doi:10.18653/v1/2022.acl-long.62 2022
[5]

Haug, D. T. T., & Jøhndal, M. (2008). Creating a parallel treebank of the Old Indo-European Bible transla- tions. In Proceedings of the Second Workshop on Language Technology for Cultural Heritage Data (pp. 27– 34)

work page 2008
[6]

Haug, D. T. T., Jøhndal, M., Eckhoff, H. M., Welo, E., Hertzenberg, M., & Müth, A. (2009). Computa- tional and linguistic issues in designing a syntactically annotated parallel corpus of Indo-European languages. Traitement Automatique des Langues , 50(2), 17–45

work page 2009
[7]

Lavidas, N. (2021). The diachrony of written language contact: A contrastive approach (Brill’s Studies in Historical Linguistics, Vol. 15). Brill

work page 2021
[8]

List, J.-M. (2014). Sequence comparison in historical linguistics (Dissertations in Language and Cognition, Vol. 1). Düsseldorf University Press. 21

work page 2014

Show all 11 references

[9]

Qi, P., Zhang, Y., Zhang, Y., Bolton, J., & Manning, C. D. (2020). Stanza: A Python natural language processing toolkit for many human languages. In Proceedings of the 58th Annual Meeting of the Associa- tion for Computational Linguistics: System Demonstrations (pp. 101–108). ...

2020 doi
[10]

M., Sander, P., Sköldberg, E., Theuer Linke, L., Zhang, T., Tahmasebi, N., Kuhn, J., & Schulte im Walde, S

Schlechtweg, D., Virk, S. M., Sander, P., Sköldberg, E., Theuer Linke, L., Zhang, T., Tahmasebi, N., Kuhn, J., & Schulte im Walde, S. (2024). The DURel annotation tool: Human and computational measurement of semantic proximity, sense clusters and semantic change. In Proceeding...

2024 doi
[11]

Sphrantzes, G. (1966). Memorii 1401–1477 (V. Grecu, Ed.; with Pseudo-Phrantzes: Macarie Melissenos, Cronica 1258–1481; Scriptores Byzantini, Vol. 5). Editura Academiei Republicii Socialiste România. 22

1966

Pith tools

Reviewed July 3, 2026 · model on record in the stance chip above.

[1] [1]

Bary, C., & Haug, D. T. T. (2011). Temporal anaphora across and inside sentences: The function of participles. Semantics and Pragmatics, 4(8), 1–56. https://doi.org/10.3765/sp.4.8

work page doi:10.3765/sp.4.8 2011

[2] [2]

Dou, Z.-Y., & Neubig, G. (2021). Word alignment by fine-tuning embeddings on parallel cor- pora. In Proceedings of the 16th Conference of the European Chapter of the Association for Com- putational Linguistics: Main Volume (pp. 2112–2128). Association for Computational Linguistics. https://doi.org/10.18653/v1/2021.eacl-main.181

work page doi:10.18653/v1/2021.eacl-main.181 2021

[3] [3]

M., Bech, K., Bouma, G., Eide, K., Haug, D

Eckhoff, H. M., Bech, K., Bouma, G., Eide, K., Haug, D. T. T., Haugen, O. E., & Jøhndal, M. (2018). The PROIEL treebank family: A standard for early attestations of Indo-European languages. Language Resources and Evaluation, 52(1), 29–65. https://doi.org/10.1007/s10579-017-9388-5

work page doi:10.1007/s10579-017-9388-5 2018

[4] [4]

Feng, F., Yang, Y., Cer, D., Arivazhagan, N., & Wang, W. (2022). Language-agnostic BERT sen- tence embedding. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (pp. 878–891). Association for Computational Linguistics. https://doi.org/10.18653/v1/2022.acl-long.62

work page doi:10.18653/v1/2022.acl-long.62 2022

[5] [5]

Haug, D. T. T., & Jøhndal, M. (2008). Creating a parallel treebank of the Old Indo-European Bible transla- tions. In Proceedings of the Second Workshop on Language Technology for Cultural Heritage Data (pp. 27– 34)

work page 2008

[6] [6]

Haug, D. T. T., Jøhndal, M., Eckhoff, H. M., Welo, E., Hertzenberg, M., & Müth, A. (2009). Computa- tional and linguistic issues in designing a syntactically annotated parallel corpus of Indo-European languages. Traitement Automatique des Langues , 50(2), 17–45

work page 2009

[7] [7]

Lavidas, N. (2021). The diachrony of written language contact: A contrastive approach (Brill’s Studies in Historical Linguistics, Vol. 15). Brill

work page 2021

[8] [8]

List, J.-M. (2014). Sequence comparison in historical linguistics (Dissertations in Language and Cognition, Vol. 1). Düsseldorf University Press. 21

work page 2014