Taylor Arnold


pdf bib
Enriching Historic Photography with Structured Data using Image Region Segmentation
Taylor Arnold | Lauren Tilton
Proceedings of the 1st International Workshop on Artificial Intelligence for Historical Image Enrichment and Access

Cultural institutions such as galleries, libraries, archives and museums continue to make commitments to large scale digitization of collections. An ongoing challenge is how to increase discovery and access through structured data and the semantic web. In this paper we describe a method for using computer vision algorithms that automatically detect regions of “stuff” — such as the sky, water, and roads — to produce rich and accurate structured data triples for describing the content of historic photography. We apply our method to a collection of 1610 documentary photographs produced in the 1930s and 1940 by the FSA-OWI division of the U.S. federal government. Manual verification of the extracted annotations yields an accuracy rate of 97.5%, compared to 70.7% for relations extracted from object detection and 31.5% for automatically generated captions. Our method also produces a rich set of features, providing more unique labels (1170) than either the captions (1040) or object detection (178) methods. We conclude by describing directions for a linguistically-focused ontology of region categories that can better enrich historical image data. Open source code and the extracted metadata from our corpus are made available as external resources.


Cross-Discourse and Multilingual Exploration of Textual Corpora with the DualNeighbors Algorithm
Taylor Arnold | Lauren Tilton
Proceedings of the Second Joint SIGHUM Workshop on Computational Linguistics for Cultural Heritage, Social Sciences, Humanities and Literature

Word choice is dependent on the cultural context of writers and their subjects. Different words are used to describe similar actions, objects, and features based on factors such as class, race, gender, geography and political affinity. Exploratory techniques based on locating and counting words may, therefore, lead to conclusions that reinforce culturally inflected boundaries. We offer a new method, the DualNeighbors algorithm, for linking thematically similar documents both within and across discursive and linguistic barriers to reveal cross-cultural connections. Qualitative and quantitative evaluations of this technique are shown as applied to two cultural datasets of interest to researchers across the humanities and social sciences. An open-source implementation of the DualNeighbors algorithm is provided to assist in its application.