# In Codice Ratio

In Codice Ratio is a research project using AI and machine learning to transcribe and analyze medieval manuscripts from the Vatican Secret Archives, focusing on historical document digitization.

In Codice Ratio is an interdisciplinary research project that applies [artificial-intelligence](https://www.wikiprompt.org/wiki/artificial-intelligence) and [machine-learning](https://www.wikiprompt.org/wiki/machine-learning) techniques to the transcription and analysis of medieval manuscripts, particularly those held in the Vatican Secret Archives. The project's name, Latin for 'in the code ratio', reflects its goal of developing computational methods to read and interpret historical documents that are often damaged, handwritten, or otherwise difficult for traditional optical character recognition systems to process.

The initiative brings together computer scientists, historians, and philologists to create automated pipelines that convert scanned images of ancient texts into searchable digital formats. By leveraging modern [deep-learning](https://www.wikiprompt.org/wiki/deep-learning) models, the project aims to unlock vast quantities of previously inaccessible archival material, enabling new historical research and preservation efforts.

## Origins and Motivation

The project emerged from the growing need to digitize large collections of handwritten historical documents. Unlike printed texts, medieval manuscripts feature irregular letterforms, abbreviations, and degradation over time, which pose significant challenges for conventional software. In Codice Ratio addresses this by training [neural-network](https://www.wikiprompt.org/wiki/neural-network) models on annotated samples of specific script styles, allowing the system to learn the visual patterns of particular scribes or periods.

The collaboration initially focused on the Vatican Secret Archives, which contain centuries of papal correspondence and administrative records. The sheer volume of material - running into millions of pages - made manual transcription impractical, motivating the development of semi-automated tools that could assist human experts rather than replace them.

## Technical Approach

At its core, In Codice Ratio employs a combination of image preprocessing, character segmentation, and sequence recognition. The pipeline typically involves several stages: first, digitized pages are cleaned and normalized to reduce noise; second, lines and individual characters are detected; and third, a [recurrent-neural-network](https://www.wikiprompt.org/wiki/recurrent-neural-network) or [transformer](https://www.wikiprompt.org/wiki/transformer)-based model transcribes the sequence of symbols into modern Latin or Italian text.

A distinctive feature is the use of synthetic data augmentation. Because labeled historical data are scarce, the team generates artificial training examples by rendering modern fonts with simulated degradation, such as ink bleed, parchment texture, and irregular spacing. This 'data-augmentation' strategy improves model robustness and reduces the need for extensive manual annotation.

The project also incorporates [curriculum-learning](https://www.wikiprompt.org/wiki/curriculum-learning), starting with simpler, cleaner documents and progressively introducing more challenging material. This staged training helps models generalize better to diverse script variants. Additionally, [beam-search](https://www.wikiprompt.org/wiki/beam-search) decoding is used during inference to produce more coherent transcription outputs by considering multiple possible character sequences.

## Key Achievements and Collaborations

In Codice Ratio has published several proof-of-concept studies demonstrating high accuracy on test sets of medieval Latin manuscripts. The team has reported character error rates below 5 percent on certain well-defined script families, a notable result for historical handwriting recognition. These findings have been presented at digital humanities conferences and in peer-reviewed venues, contributing to a growing literature on computational paleography.

The project maintains active partnerships with institutions such as the Vatican School of Palaeography and various European universities. Researchers involved include computer scientists specializing in computer vision and natural language processing, as well as medieval historians who provide domain expertise for annotation and validation. The collaboration has also explored extensions to other archival collections, including early modern legal documents and Renaissance humanist letters.

## Impact on Digital Humanities

By demonstrating that modern AI methods can be adapted to ancient scripts, In Codice Ratio has influenced broader digital humanities initiatives. Its techniques for handling noisy, handwritten input are relevant to archives worldwide, and the project has released portions of its annotated datasets and model code under open licenses to encourage replication and further research.

Scholars have used the project's outputs to support studies on papal diplomacy, economic history, and linguistic evolution. The ability to search and cross-reference millions of transcribed documents opens new avenues for quantitative historical analysis, a field that is still in its infancy.

## Current Status and Future Directions

As of the early 2020s, In Codice Ratio continues to refine its models, with ongoing work on integrating [large-language-model](https://www.wikiprompt.org/wiki/large-language-model) components for context-aware correction and semantic search. The team is also investigating unsupervised and weakly supervised methods to reduce annotation costs further.

Future plans include expanding to other script families, such as Gothic and Carolingian minuscule, and developing user-friendly interfaces for scholars who are not technical specialists. The project also aims to establish standards for evaluating handwriting recognition systems in historical contexts, which would help benchmark progress across different research groups.

Given the rapid pace of AI development, the project is well-positioned to incorporate advances from [generative-ai](https://www.wikiprompt.org/wiki/generative-ai) and [transformer](https://www.wikiprompt.org/wiki/transformer) architectures, potentially enabling end-to-end transcription of entire documents without manual segmentation. Such breakthroughs would significantly accelerate the digitization of global archival heritage.

---
Source: https://www.wikiprompt.org/wiki/in-codice-ratio
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-14T06:30:48.461335+00:00