Published Nov 27, 2019

97 - Automated Analysis Of Historical Printed Documents, With Taylor Berg-Kirkpatrick

Taylor Berg-Kirkpatrick delves into the cutting-edge intersection of machine learning and historical document analysis, highlighting the challenges and innovations in OCR for historical texts, uncovering printing biases, and using novel unsupervised learning techniques to reveal hidden insights from the past.
Episode Highlights
NLP Highlights logo

Popular Clips

Episode Highlights

  • Compositor Techniques

    Compositor attribution techniques are crucial for understanding the historical context of printed texts. explains that identifying the original compositors of Shakespeare's works can reveal biases and errors introduced during printing, which affect the authenticity of the texts 1. By using unsupervised learning models, researchers can analyze historical documents without relying on modern supervised data, treating the OCR problem as a decipherment challenge 2. This approach allows for a more accurate reconstruction of historical texts, as Taylor notes:

    You try to fit a model that treats the text as a latent variable.

    ---

    These techniques not only enhance our understanding of historical documents but also preserve the integrity of literary works.

       

    Printing Identification

    Printing house identification is another fascinating aspect of historical document analysis. Taylor highlights the example of John Milton's "Areopagitica," a pamphlet printed secretly to avoid persecution, illustrating the challenges of attributing such works to specific printing houses 3. By examining damaged glyphs and using AI, researchers can link clandestine documents to known printers, offering insights into the political and historical context of the time 4. Taylor describes the process as akin to manual detective work, emphasizing the potential of AI to scale this analysis:

    Can you use AI and machine learning to scale that same reasoning process?

    ---

    This method not only aids in historical research but also underscores the intersection of technology and humanities.