Word Vectors in the Eighteenth Century

Word Vectors

This page is meant as a set of links and resources related to my work using word vectors to study eighteenth-century literature. This work asks the question: how can new vector-based models of semantics reveal the historicity of specific configurations of meaning in eighteenth-century literature? Most of this work is published serially as blog posts, linked below. The later of these are “slideshow essays”-experiments with the forms of visual rhetoric that work so well in the digital humanities-rather than traditional blog posts. There is also a video of a talk I’ve given about this work. Lastly, I’ve uploaded several word2vec models I’m using, trained on a corpus of eighteenth-century literature; and linked to some relevant code (more code will be coming soon).

Posts

• Episode 1: Concepts (14 Apr 2016)
• Episode 2: Methods (1 Jun 2016)
• Episode 3: From Fields to Vectors (10 Sep 2016)
• Episode 4: Semantic Networks (25 Sep 2016)

Video

• Presentation at IPAM in UCLA (25 May 2016)

Appendices

• “Would Word2Vec get into grad school? Evaluating Word2Vec against the Miller Analogies Test”

Models

• Word2Vec Models for Twenty-year Periods of 18C (ECCO, “Literature and Language,” 1700-99) (150 million words each; skip-gram size of 10 words)

Code

• Code to evaluate a word2vec model against the Miller Analogies Test
• Code to produce a semantic network from a gensim word2vec model
• Code for aligning two gensim word2vec models using Procrustes matrix alignment