Our approach builds on these studies by interpreting the dimensions of embedding models as representative of meaningful cul-tural categories rather than simply biases, distortions, or deficits in the semantic system. We then use these dimensions as tools to illu-minate complex cultural relations associated with class in a given social context, across contexts, and over time. More broadly, this article is the first to specifically demonstrate the utility of word embedding models for sociological and cultural inquiry.

Word embedding algorithms input large collections of digitized text and output a high-dimensional vector-space model1 in which each unique word is represented as a vector in the space (Mikolov, Yih, and Zweig 2013; Pennington, Socher, and Manning 2014). This means each word appearing in the analyzed documents is ascribed a set of coordinates that fix its loca-tion in a geometric space in relation to every other word. Words are positioned in this space based on their surrounding “context” words in the text, such that words sharing many con-texts are positioned near one another, and words that inhabit different linguistic contexts are located farther apart. Previous work with word embeddings in computational linguistics shows that words frequently sharing contexts, and thus located nearby in the vector space, tend to share similar meanings

Either the PPMI model or the tf-idf model can be used to compute word simi-larity, for tasks like finding word paraphrases, tracking changes in word meaning, orautomatically discovering meanings of words in different corpora. For example, wecan find the 10 most similar words to any target wordwby computing the cosinesbetweenwand each of theV−1 other words, sorting, and looking at the top 10.

In summary, the vector semantics model we’ve described so far represents a targetword as a vector with dimensions corresponding either to to the documents in a largecollection (the term-document matrix) or to the counts of words in some neighboringwindow (the term-term matrix). The values in each dimension are counts, weightedby tf-idf (for term-document matrices) or PPMI (for term-term matrices), and thevectors are sparse (since most values are zero)

A (rough) graphical demonstration of cosine similarity, showing vectors forthree words (cherry,digital, andinformation) in the two dimensional space defined by countsof the wordscomputerandpienearby. Note that the angle betweendigitalandinformationissmaller than the angle betweencherryandinformation. When two vectors are more similar,the cosine is larger but the angle is smaller; the cosine has its maximum (1) when the anglebetween two vectors is smallest (0◦); the cosine of all other angles is less than 1

We will see that this method results in very longvectors that aresparse, i.e. mostly zeros (since most words simply never occur inthe context of others).

a thin blue column of smoke rose.

The house itself was of timbers   Hewn from the cypress-tree, and carefully fitted together.   Large and low was the roof;

On the banks of the Têche are the towns of St. Maur and St. Martin.   There the long-wandering bride shall be given again to her bridegroom,   There the long-absent pastor regain his flock and his sheepfold.   Beautiful is the land, with its prairies and forests of fruit-trees;   Under the feet a garden of flowers, and the bluest of heavens   Bending above, and resting its dome on the walls of the forest.   They who dwell there have named it the Eden of Louisiana."

Silent at times, then singing familiar Canadian boat-songs,   Such as they sang of old on their own Acadian rivers,   And through the night were heard the mysterious sounds of the desert,   Far off,—indistinct,—as of wave or wind in the forest,   Mixed with the whoop of the crane and the roar of the grim alligator.

we never have sworn them allegiance!

Michael the fiddler was placed, with the gayest of hearts and of waistcoats.

Merrily, merrily whirled the wheels of the dizzying dances   Under the orchard-trees and down the path to the meadows;   Old folk and young together, and children mingled among them.   Fairest of all the maids was Evangeline, Benedict's daughter!   Noblest of all the youths was Gabriel, son of the blacksmith!

"Safer are we unarmed, in the midst of our flocks and our cornfields,   Safer within these peaceful dikes, besieged by the ocean,   Than our fathers in forts, besieged by the enemy's cannon.

There are, one should note, many of the latter, but they always seem about to drown in the shrill orthographical chaos surrounding them, complaints often written by those who look forward to the demise of critics — and editors — with a populist glee.

Each of these approaches—data as text, artifact, and processable information—allow one to produce or uncover evidence that can support particular claims and arguments. Data is not in and of itself a kind of evidence but a multifaced object which can be mobilized as evidence in support of an argument.

Cavalcade of the American Negro