Aeneas transforms the way historians connect the past

The introduction of the first model of contextualization of ancient inscriptions designed to help historians in better interpretation, assignment and restoration of fragmentary texts.

Writing was everywhere in the Roman world – engraved on everything, from imperial monuments to everyday objects. From political graffiti, love poems and epitaphs to business transactions, birthday invitations and magical spells, inscriptions are offered by contemporary historians, rich insight into the variety of everyday life throughout the Roman world.

Often these texts are fragmentary, weathered or intentionally damaged. Restoring, dating and placing them is almost impossible without contextual information, especially when comparing similar inscriptions.

Today we publish paper Introduction in nature AeneasThe first model of artificial intelligence (AI) to contextualize ancient subtitles.

While working with ancient subtitles, historians traditionally rely on their specialist knowledge and specialized resources to identify “similarities” – which are texts that divide similarities in formulation, syntax, standardized formulas or origin.

Aeneas accelerates this complex and time -consuming work very much. This is a reason for thousands of Latin inscriptions, recovering text and contextual similarities in a few seconds that allow historians to interpret and rely on the model's arrangements.

Our model can also be adapted to other ancient languages, scripts and media, from papyrus to coins, expanding your capabilities to help in the use of connections in a wider scope of historical evidence.

We developed Aeneas from the University of Nottingham and in cooperation with scientists at the universities of Warwick, Oxford and Athena of the University of Economics and Business (AUEB). This work was part of a broader effort to examine how generative AI can help historians better identify and interpret similarities on a large scale.

We want these research to benefit as many people as possible, which is why we create an interactive version of Aeneas freely available to researchers, students, teachers, specialists from museums and others PredictingThePast.com. To support further research, we are also open sourcing Our code and data set.

Aeneasa advanced capabilities

Named after the wandering hero of Graec-Roman mythology, Aeneas is based on ithaki, our previous work using artificial intelligence to restore, date and place ancient Greek inscriptions.

Aeneas goes a step further, helping historians interpret and contextualize the text, giving meaning to isolated fragments, draw richer conclusions and combine a better understanding of ancient history.

Advanced capabilities of our model include:

• Parallels Search: He is looking for similarities in a wide collection of Latin inscriptions. By changing each text into a kind of historical fingerprint, AEneas identifies deep connections that can help historians in the situation of subtitles in their wider historical context.
• Processing of multimodal input data: Aeneas is the first model that determines the geographical origin of the text using multimodal inputs. Analyzes both text and visual information, such as inscription images.
• Restoring gaps of unknown length: For the first time, Aeneas can restore gaps in texts in which the missing length is unknown. This makes it a more versatile tool for historians dealing with heavily damaged material.
• The most modern performance: Aeneas sets the new most modern reference point in restoring damaged texts and predictions when and where they were written.

How aeneas works

Aeneas is a multimodal generative neural network that adopts the text and image of the inscription as input data. To train Aeneas, we curled a large and reliable set of data, drawing on decades of historians' work to create digital collections, especially Roma epigraphic database (EDR), Heidelberg epigraphic database (Edh) i Epigraphic Clauss database (EDCS-ELT).

We cleaned, harmonized and combined these records with one set of data that can be found, we call Latin -epigraphic data set (LED), covering over 176,000 Latin inscriptions from all over the ancient Roman world.

Our model uses a decoder based on the transformer for the text processing of the inscription input. Specialized networks support the restoration and dating of characters using the text, while geographical attribution also uses inscriptions images as inputs. The decoder downloads similar LED subtitles, ordered by meaning.

In the case of each inscription, the Aeneas contextualization mechanism regains a list of similarities using a technique called “embedding” – coding the text and contextual information of each inscription in a type of historical fingerprint, containing details of what the text says, his language, when and where it comes, and where it comes from, and how he refers to other subtitles.

The most modern results

Aeneas groups inscriptions by date of writing much more clearly than other general purpose models, also trained in Latin, as shown in the visualization below.

Aeneas restores damaged inscriptions with the highest accuracy of 73% in gaps up to ten characters. This only decreases to 58%when the restore length is unknown – in itself an extremely difficult task. It also shows his reasoning in an interpretative manner, providing maps of significance that emphasize which parts of the input data influenced its forecasts. By using visual data, our model can assign the inscription of one of 62 ancient Roman provinces with 72% accuracy. On dates, Aeneas places a SMS -in 13 years from the scope of dates delivered by historians.

A new lens on historical debates

To test Aeneas's capabilities during the ongoing research debate, we gave her one of the most famous Roman inscriptions: August's achievements Emperor August from the first person with his achievements.

Historians argue for a long time the dates of this inscription. Instead of predicting a single permanent date, Aenea created a detailed distribution of possible dates, showing two separate peaks, with one smaller peak of about 10-1 BC and a larger, more reliable quib between 10-20 ne results, these results captured both prevailing dating hypotheses in a quantitative way.

Aeneas based his forecasts on subtle language features and historical markers, such as official titles and monuments listed in the text. By changing the dating question into a probabilistic estimate based on language and contextual data, our model offers a new, quantitative way of engaging in long -term historical debates.

Most importantly, Aeneas also regained many important similarities from the imperial legal texts related to August's heritage, emphasizing how the ideology of the Empire in the media and geography was represented.

Developing historical research

To assess the influence of Aeneas as a help on research, we conducted a large -scale historian and joint study AI. We invited twenty -three historians who work regularly with subtitles to restore, date and put a set of texts using Aeneas.

Our assessment, summed up in the table below, shows how the most effective results were achieved when historians used AEneas contextual information along with its predictions of restoring and assigning Roman inscriptions.

Aeneas has helped historians to identify new similarities in our study and increased their confidence during the solution to complex epigraphic tasks. Historians consistently emphasized the value of Aeneas in accelerating their work and expanding the scope of the most appropriate parallel inscriptions.

Sharing tools, shaping the future

Aeneas aims to integrate with the existing flow of the work of historians. Combining expert knowledge with machine learning, he opens the cooperation process, offering interpretative suggestions that serve as valuable starting points for historical research.

As part of today's edition, we update Itaka, our ancient Greek model, which will be powered by AEneas and contain a contextualization function, supplementing unknown length and generally better performance.

We also worked on new ones Teaching the curriculum to bridge technical skills with historical thinking in the classroom. This program is consistent with the initiatives related to AI, including the European Commission Digital competence framework for citizens (DigComp 2.2), UNESCO AI competences for studentsand the announcement of the European Commission and the Organization of Economic Cooperation and Development (OECD) Ailil Framework.

The Aeneas team continues to cooperate with various subject experts, using Aeneas to help shed light at our ancient past – with more.