A Probabilistic Generative Model for Typographical Analysis of Early Modern Printing

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 2954–2960 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 2954 A Probabilistic Generative Model for Typographical Analysis of Early Modern Printing Kartik Goyal1 Chris Dyer2 Christopher Warren3 Max G’Sell4 Taylor Berg-Kirkpatrick5 1Language Technologies Institute, Carnegie Mellon University 3Department of English, Carnegie Mellon University2Deepmind 4Department of Statistics, Carnegie Mellon University 5Computer Science and Engineering, University of California, San Diego {kartikgo,cnwarren,mgsell}@andrew.cmu.edu cdyer@google.com tberg@eng.ucsd.edu Abstract We propose a deep and interpretable prob- abilistic generative model to analyze glyph shapes in printed Early Modern documents. We focus on clustering extracted glyph images into underlying templates in the presence of multiple confounding sources of variance. Our approach introduces a neural editor model that first generates well-understood printing phe- nomena like spatial perturbations from tem- plate parameters via interpertable latent vari- ables, and then modifies the result by generat- ing a non-interpretable latent vector responsi- ble for inking variations, jitter, noise from the archiving process, and other unforeseen phe- nomena associated with Early Modern print- ing. Critically, by introducing an inference network whose input is restricted to the vi- sual residual between the observation and the interpretably-modified template, we are able to control and isolate what the vector-valued latent variable captures. We show that our approach outperforms rigid interpretable clus- tering baselines (Ocular) and overly-flexible deep generative models (VAE) alike on the task of completely unsupervised discovery of typefaces in mixed-font documents. 1 Introduction Scholars interested in understanding details related to production and provenance of historical docu- ments rely on methods of analysis ranging from the study of orthographic differences and stylometrics, to visual analysis of layout, font, and printed char- acters. Recently developed tools like Ocular (Berg- Kirkpatrick et al., 2013) for OCR of historical docu- ments have helped automate and scale some textual analysis methods for tasks like compositor attri- bution (Ryskina et al., 2017) and digitization of historical documents (Garrette et al., 2015). How- ever, researchers often find the need to go beyond Figure 1: We desire a generative model that can be biased to cluster according to typeface characteristics (e.g. the length of the middle arm) rather than other more visually salient sources of variation like inking. textual analysis for establishing provenance of his- torical documents. For example, Hinman (1963)’s study of typesetting in Shakespeare’s First Folio relied on the discovery of pieces of damaged or distinctive type through manual inspection of every glyph in the document. More recently, Warren et al. (2020) examine pieces of distinctive types across several printers of the early modern period to posit the identity of clandestine printers of John Milton’s Areopagitica (1644). In such work, researchers frequently aim to determine whether a book was produced by a single or multiple printers (Weiss (1992); Malcolm (2014); Takano (2016)). Hence, in order to aid these visual methods of analyses, we propose here a novel probabilistic generative model for analyzing extracted images of individ- ual printed characters in historical documents. We draw from work …

Content truncated to 500 words