Multimodal Text Analysis of Medieval Illuminated Manuscripts

In this challenge, we delve into the realm of historical document analysis. Your task is to develop an AI system that can extract relevant information and insights from medieval illuminated manuscripts. The twist: these manuscripts are multimodal, containing both text and intricate illustrations.

Constraints

1. Dataset: You will be provided with a collection of scanned medieval illuminated manuscripts, each consisting of multiple pages with text and illustrations. The manuscripts date back to the 12th to 15th centuries and belong to various European styles.
2. Text Complexity: The text within the manuscripts is primarily written in Latin, with occasional use of vernacular languages (e.g., Old French, Middle English). The text can be quite complex, featuring cursive script, elaborate punctuation, and variable line spacing.
3. Illustration Recognition: The illustrations within the manuscripts are intricate and can range from simple ornaments to elaborate scenes. Your system should be able to recognize and interpret these illustrations, understanding their role in conveying meaning and context.
4. Information Extraction: Your AI system must be able to extract relevant information from both the text and the illustrations, including:

• Key events, people, and locations mentioned in the text
• Symbols and motifs used in the illustrations and their corresponding meanings
• Relationships between text and illustrations, such as which illustrations accompany specific paragraphs or sections of text

1. Contextual Understanding: Your system should be able to contextualize the extracted information, recognizing how it relates to the broader historical and cultural context of the manuscripts.
2. Interpretability: Since the manuscripts are a valuable historical resource, your system must be transparent and explainable in its decision-making process. This includes providing clear reasoning behind the extracted information and its relevance to the context.

Evaluation Metrics

Your system will be evaluated on:

1. Accuracy of information extraction (text and illustration-based)
2. Quality of contextual understanding and interpretation
3. Interoperability with other historical documents and resources

Timeline

The challenge will run for 6 weeks. You will have access to the dataset from week 1 to week 4, during which time you will develop and train your system. In week 5, you will submit a written report and a demo of your system. In week 6, we will evaluate your submissions and provide feedback.

Prizes and Recognition

The participant who demonstrates the most comprehensive and accurate solution will receive recognition and a prize. Additional prizes will be awarded for notable achievements in specific areas, such as illustration recognition or contextual understanding.

What will you bring to the challenge?

Publicado automáticamente