---
title: "Ithaca Revisited — Locaputo 2026"
description: "Epigraphers working with damaged Greek inscriptions must fill in lost characters and decide where and when a text was produced."
authors: ["Alessandro Locaputo", "Andrea Brunello", "Nicola Saccomanno", "Paraskevi Platanou", "Giuseppe Serra"]
year: 2026
venue: "Proceedings of the Fifteenth Language Resources and Evaluation Conference"
source: "https://aclanthology.org/2026.lrec-1.82/"
doi: "10.63317/3gucgvmwsf45"
licence: "CC BY-NC 4.0 (ELRA, LREC proceedings)"
licence_url: "https://creativecommons.org/licenses/by-nc/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/papers/locaputo-2026-ithaca-revisited-benchmarking-domain-specific/"
---
[Elamur Research](https://md.research.elamur.ai/) · [this page in HTML](https://research.elamur.ai/free/papers/locaputo-2026-ithaca-revisited-benchmarking-domain-specific/)

[home](https://md.research.elamur.ai/) / [papers](https://md.research.elamur.ai/free/papers/) / [2026](https://md.research.elamur.ai/free/papers/2026/page/3/) / Ithaca Revisited: Benchmarking a Domain-Specific Model for Epigraphy in the Age of LLMs

# Ithaca Revisited: Benchmarking a Domain-Specific Model for Epigraphy in the Age of LLMs

**Authors:** Alessandro Locaputo, Andrea Brunello, Nicola Saccomanno, Paraskevi Platanou, Giuseppe Serra · **Year:** 2026 · **Venue:** Proceedings of the Fifteenth Language Resources and Evaluation Conference · **Licence:** [CC BY-NC 4.0 (ELRA, LREC proceedings)](https://creativecommons.org/licenses/by-nc/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.lrec-1.82/) · [PDF](https://aclanthology.org/2026.lrec-1.82.pdf) · [DOI 10.63317/3gucgvmwsf45](https://doi.org/10.63317/3gucgvmwsf45) · [OpenAlex](https://openalex.org/W7160160040)

**Citations:** 0 citations (openalex-snapshot · 2026-10-14)

## Summary
Epigraphers working with damaged Greek inscriptions must fill in lost characters and decide where and when a text was produced. Ithaca, a Transformer model released in 2022 and trained on the I.PHI corpus, became a standard tool for these tasks. The rise of general-purpose LLMs raises the question of whether such a specialised system is still needed. The authors address this in two ways. First, they compare Ithaca with GPT-5 on restoration, geographic attribution and dating. GPT-5 is supplied with retrieved related inscriptions as in-context examples. Second, they profile Ithaca's behaviour, measuring how accuracy changes with gap length and placement, region of origin and subject matter. Topics are derived through GSDMM clustering, and differences are tested statistically. Ithaca still clearly beats GPT-5. The analysis also shows how over-represented regions and themes in the training data shape its strengths and failures. The study suggests domain-specific models keep an edge in data-poor humanities fields. It also gives researchers and teachers a practical map of when Ithaca can be trusted.

## Key points
- The baseline is GPT-5 using retrieval-augmented in-context learning. Ithaca substantially outperforms it on all three core tasks (restoration, place attribution and date attribution).
- I.PHI, a cleaned and normalised derivative of the PHI collection (about 180k texts), contains 78,608 machine-actionable Greek inscriptions. The analyses use a test set of 7,127 inscriptions.
- Ithaca combines character- and word-level embeddings in a shared Transformer torso with three shallow task heads. It chooses among 84 regions and dates texts in 10-year bins spanning 800 BCE–800 CE.
- The study is organised around six research questions (RQ1–RQ6). The factors it investigates are how long lacunae are and where they fall in the text, the provenance of each inscription, and its semantic theme.
- Topics come from GSDMM tuned by coherence-based grid search (α = 0.3, β = 0.05, k = 10). An additional "Unknown" class holds inscriptions whose topic probability is ≤ 0.7.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

Fields: [Linguistics & languages](https://md.research.elamur.ai/free/shelves/linguistics/)

## Cite

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{locaputo2026ithaca,
  title = {{Ithaca Revisited: Benchmarking a Domain-Specific Model for Epigraphy in the Age of LLMs}},
  author = {Locaputo, Alessandro and Brunello, Andrea and Saccomanno, Nicola and Platanou, Paraskevi and Serra, Giuseppe},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/3gucgvmwsf45},
  url = {https://doi.org/10.63317/3gucgvmwsf45}
}
```

- [← previous Is This Idea Novel? An Automated Benchmark for Judgment of Research Ideas](https://md.research.elamur.ai/free/papers/schopf-2026-is-this-idea-novel-automated/)
- [next → ITUNLP at MWE-2026 AdMIRe 2: A Zero-Shot LLM Pipeline for Multimodal Idiom Understanding and Ranking](https://md.research.elamur.ai/free/papers/site-2026-itunlp-mwe-2026-admire-2/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/papers/2026/page/3/)

## References

22 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Yannis Assael, Thea Sommerschield, Alison Cooley, Brendan Shillingford, John Pavlopoulos, Priyanka Suresh, Bailey Herms, Justin Grayston, Benjamin Maynard, Nicholas Dietrich, et al. 2025. Contextualizing ancient texts with generative neural networks. Nature, pages 1–7. [find](https://scholar.google.com/scholar?q=Yannis+Assael%2C+Thea+Sommerschield%2C+Alison+Cooley%2C+Brendan+Shillingford%2C+John+Pavlopoulos%2C+Priyanka+Suresh%2C+Bailey+Herms%2C+Justin+Grayston%2C+Benjamin+Maynard%2C+Nicholas+Dietrich%2C+et+al%2E+2025%2E+Contextualiz)
2. Yannis Assael, Thea Sommerschield, and Jonathan Prag. 2019. Restoring ancient text using deep learning: A case study on Greek epigraphy. arXiv preprint arXiv:1910.06262. [find](https://scholar.google.com/scholar?q=Yannis+Assael%2C+Thea+Sommerschield%2C+and+Jonathan+Prag%2E+2019%2E+Restoring+ancient+text+using+deep+learning%3A+A+case+study+on+Greek+epigraphy%2E+arXiv+preprint+arXiv%3A1910%2E06262%2E)
3. Yannis Assael, Thea Sommerschield, Brendan Shillingford, Mahyar Bordbar, John Pavlopoulos, Marita Chatzipanagiotou, Ion Androutsopoulos, Jonathan Prag, and Nando De Freitas. 2022. Restoring and attributing ancient texts using deep neural networks. Nature, 603(7900):280– 283. [find](https://scholar.google.com/scholar?q=Yannis+Assael%2C+Thea+Sommerschield%2C+Brendan+Shillingford%2C+Mahyar+Bordbar%2C+John+Pavlopoulos%2C+Marita+Chatzipanagiotou%2C+Ion+Androutsopoulos%2C+Jonathan+Prag%2C+and+Nando+De+Freitas%2E+2022%2E+Restoring+and+attrib)
4. David Bamman and Patrick J Burns. 2020. Latin BERT: A contextual language model for classical philology. arXiv preprint arXiv:2009.10053. — [arXiv:2009.10053](https://arxiv.org/abs/2009.10053) [find](https://scholar.google.com/scholar?q=David+Bamman+and+Patrick+J+Burns%2E+2020%2E+Latin+BERT%3A+A+contextual+language+model+for+classical+philology%2E+arXiv+preprint+arXiv%3A2009%2E10053%2E+%E2%80%94+%5BarXiv%3A2009%2E10053%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F2009%2E10053%29)
5. David M Blei, Andrew Y Ng, and Michael I Jordan. 2003. Latent dirichlet allocation. Journal of machine Learning research, 3(Jan):993–1022. [find](https://scholar.google.com/scholar?q=David+M+Blei%2C+Andrew+Y+Ng%2C+and+Michael+I+Jordan%2E+2003%2E+Latent+dirichlet+allocation%2E+Journal+of+machine+Learning+research%2C+3%28Jan%29%3A993%E2%80%931022%2E) [our card](https://md.research.elamur.ai/free/papers/blei-2003-latent-dirichlet-allocation/)
6. Andrea Brunello, Emanuela Colombi, Alessandro Locaputo, Stefano Magnani, Nicola Saccomanno, Giuseppe Serra, et al. 2023. Usage of language model for the filling of lacunae in ancient Latin inscriptions: A case study. In CEUR WORKSHOP PROCEEDINGS, volume 3536, pages 113–125. CEUR-WS. [find](https://scholar.google.com/scholar?q=Andrea+Brunello%2C+Emanuela+Colombi%2C+Alessandro+Locaputo%2C+Stefano+Magnani%2C+Nicola+Saccomanno%2C+Giuseppe+Serra%2C+et+al%2E+2023%2E+Usage+of+language+model+for+the+filling+of+lacunae+in+ancient+Latin+inscription)
7. A. Buonopane. 2009. Manuale di epigrafia latina. Beni culturali. Carocci, Roma. Tex.lccn: 2009478450. [find](https://scholar.google.com/scholar?q=A%2E+Buonopane%2E+2009%2E+Manuale+di+epigrafia+latina%2E+Beni+culturali%2E+Carocci%2C+Roma%2E+Tex%2Elccn%3A+2009478450%2E)
8. Eric Cullhed. 2024. Instruct-tuning pretrained causal language models for ancient Greek papyrology and epigraphy. arXiv preprint arXiv:2409.13870. — [arXiv:2409.13870](https://arxiv.org/abs/2409.13870) [find](https://scholar.google.com/scholar?q=Eric+Cullhed%2E+2024%2E+Instruct%2Dtuning+pretrained+causal+language+models+for+ancient+Greek+papyrology+and+epigraphy%2E+arXiv+preprint+arXiv%3A2409%2E13870%2E+%E2%80%94+%5BarXiv%3A2409%2E13870%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F2409%2E13870)
9. DARIAH-Teach Project. 2022. Digital Greek and Latin epigraphy course: Test 1.10.1 Ithaca. [https://teach.dariah.eu/course/view.php?id=80](https://teach.dariah.eu/course/view.php?id=80). [↗](https://teach.dariah.eu/course/view.php?id=80)
10. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies,… [find](https://scholar.google.com/scholar?q=Jacob+Devlin%2C+Ming%2DWei+Chang%2C+Kenton+Lee%2C+and+Kristina+Toutanova%2E+2019%2E+BERT%3A+Pre%2Dtraining+of+deep+bidirectional+transformers+for+language+understanding%2E+In+Proceedings+of+the+2019+conference+of+the+N)
11. Google DeepMind. 2022. Predicting the past - Ithaca interactive tool. [https://predictingthepast.com/](https://predictingthepast.com/). Interactive web application for inscription restoration and spatio-temporal attribution. [↗](https://predictingthepast.com/)
12. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al. 2024. The Llama 3 herd of models. arXiv preprint arXiv:2407.21783. [find](https://scholar.google.com/scholar?q=Aaron+Grattafiori%2C+Abhimanyu+Dubey%2C+Abhinav+Jauhri%2C+Abhinav+Pandey%2C+Abhishek+Kadian%2C+Ahmad+Al%2DDahle%2C+Aiesha+Letman%2C+Akhil+Mathur%2C+Alan+Schelten%2C+Alex+Vaughan%2C+et+al%2E+2024%2E+The+Llama+3+herd+of+models%2E+)
13. Paul A. Iversen. 2007. The Packard Humanities Institute - Greek epigraphy project and the revolution in Greek epigraphy. Abgadiyat, 2:51–55. [find](https://scholar.google.com/scholar?q=Paul+A%2E+Iversen%2E+2007%2E+The+Packard+Humanities+Institute+%2D+Greek+epigraphy+project+and+the+revolution+in+Greek+epigraphy%2E+Abgadiyat%2C+2%3A51%E2%80%9355%2E)
14. Packard Humanities Institute. 2005. The packard humanities institute’s searchable greek inscriptions. [https://inscriptions.packhum.org/](https://inscriptions.packhum.org/). [↗](https://inscriptions.packhum.org/)
15. Peng Qi, Yuhao Zhang, Yuhui Zhang, Jason Bolton, and Christopher D Manning. 2020. Stanza: A Python natural language processing toolkit for many human languages. arXiv preprint arXiv:2003.07082. [find](https://scholar.google.com/scholar?q=Peng+Qi%2C+Yuhao+Zhang%2C+Yuhui+Zhang%2C+Jason+Bolton%2C+and+Christopher+D+Manning%2E+2020%2E+Stanza%3A+A+Python+natural+language+processing+toolkit+for+many+human+languages%2E+arXiv+preprint+arXiv%3A2003%2E07082%2E) [our card](https://md.research.elamur.ai/free/papers/ud-2020-qi-stanza-python-natural-language-processing/)
16. Rajesh PN Rao, Nisha Yadav, Mayank N Vahia, Hrishikesh Joglekar, Ronojoy Adhikari, and Iravatham Mahadevan. 2009. A markov model of the Indus script. Proceedings of the National Academy of Sciences, 106(33):13685–13690. [find](https://scholar.google.com/scholar?q=Rajesh+PN+Rao%2C+Nisha+Yadav%2C+Mayank+N+Vahia%2C+Hrishikesh+Joglekar%2C+Ronojoy+Adhikari%2C+and+Iravatham+Mahadevan%2E+2009%2E+A+markov+model+of+the+Indus+script%2E+Proceedings+of+the+National+Academy+of+Sciences%2C+1)
17. Thea Sommerschield*, Yannis Assael*, Brendan Shillingford, Mahyar Bordbar, John Pavlopoulos, Marita Chatzipanagiotou, Ion Androutsopoulos, Jonathan Prag, and Nando de Freitas. 2021. I.PHI dataset: ancient greek inscriptions. [https://github.com/sommerschield/iphi](https://github.com/sommerschield/iphi). [↗](https://github.com/sommerschield/iphi)
18. Robbe Wulgaert. 2023. Ithaca ai meets ancient Greek: Muses and robots in the classroom. Teaching History, 57(3):16–20. [find](https://scholar.google.com/scholar?q=Robbe+Wulgaert%2E+2023%2E+Ithaca+ai+meets+ancient+Greek%3A+Muses+and+robots+in+the+classroom%2E+Teaching+History%2C+57%283%29%3A16%E2%80%9320%2E)
19. Qizhe Xie, Guokun Lai, Zihang Dai, and Eduard Hovy. 2017. Large-scale cloze test dataset created by teachers. arXiv preprint arXiv:1711.03225. [find](https://scholar.google.com/scholar?q=Qizhe+Xie%2C+Guokun+Lai%2C+Zihang+Dai%2C+and+Eduard+Hovy%2E+2017%2E+Large%2Dscale+cloze+test+dataset+created+by+teachers%2E+arXiv+preprint+arXiv%3A1711%2E03225%2E)
20. Nisha Yadav, Hrishikesh Joglekar, Rajesh PN Rao, Mayank N Vahia, Ronojoy Adhikari, and Iravatham Mahadevan. 2010. Statistical analysis of the Indus script using n-grams. PLoS One, 5(3):e9506. [find](https://scholar.google.com/scholar?q=Nisha+Yadav%2C+Hrishikesh+Joglekar%2C+Rajesh+PN+Rao%2C+Mayank+N+Vahia%2C+Ronojoy+Adhikari%2C+and+Iravatham+Mahadevan%2E+2010%2E+Statistical+analysis+of+the+Indus+script+using+n%2Dgrams%2E+PLoS+One%2C+5%283%29%3Ae9506%2E)
21. Jianhua Yin and Jianyong Wang. 2014. A dirichlet multinomial mixture model-based approach for short text clustering. In Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining, pages 233–242. [find](https://scholar.google.com/scholar?q=Jianhua+Yin+and+Jianyong+Wang%2E+2014%2E+A+dirichlet+multinomial+mixture+model%2Dbased+approach+for+short+text+clustering%2E+In+Proceedings+of+the+20th+ACM+SIGKDD+international+conference+on+Knowledge+discove)
22. Yanzhao Zhang, Mingxin Li, Dingkun Long, Xin Zhang, Huan Lin, Baosong Yang, Pengjun Xie, An Yang, Dayiheng Liu, Junyang Lin, Fei Huang, and Jingren Zhou. 2025. Qwen3 embedding: Advancing text embedding and reranking through foundation models. arXiv preprint arXiv:2506.05176. [find](https://scholar.google.com/scholar?q=Yanzhao+Zhang%2C+Mingxin+Li%2C+Dingkun+Long%2C+Xin+Zhang%2C+Huan+Lin%2C+Baosong+Yang%2C+Pengjun+Xie%2C+An+Yang%2C+Dayiheng+Liu%2C+Junyang+Lin%2C+Fei+Huang%2C+and+Jingren+Zhou%2E+2025%2E+Qwen3+embedding%3A+Advancing+text+embeddin) [our card](https://md.research.elamur.ai/free/papers/zhang-2025-qwen3-embedding-advancing-text/)