Tool example: Cinemetrics (Frederic Brodbeck’s project for “measuring and visualizing movie data, in order to reveal the characteristics of films and to create a visual ‘fingerprint’ for them. Information such as the editing structure, color, speech or motion are extracted, analyzed and transformed into graphic representations so that movies can be seen as a whole and easily interpreted or compared side by side”; includes downloadable code for Python script tools used to create the metrics)
Text Analysis (recap)
Alan’s slides on relation between text-encoding and text-anaylsis (using Yin Liu’s table of paradigms of text in “Ways of Reading, Models for Text, and the Usefulness of Dead People”)
Richard Jean So and Edwin Roland, “Race and Distant Reading” (2020) — Figure 1, p. 67.
Here, we decided to focus on the postwar American novel. We constructed a corpus that is evenly split between novels by black authors and those by white authors. (61)
Our model includes three types of features: diction, syntax, and narrative. First, our model simply counts how many times certain words appear in a text…. Second, our model counts how often various parts of speech, like nouns and adverbs, appear in a text…. Finally, the model computes statistics relevant to the represented world of the story. These include the ratio of dialogue to narration, the number of characters in the story, the average amount of narrative attention paid to each character, the ratio of manufactured to natural objects in the novel, and the average number of locations or settings in it. We refer to this third category of features as “narration.” (63)
The outlines of a story emerge: novels by black authors tend to use more verbs and nouns (i.e., action and object- based language) than novels by white authors. And the latter tend to use more qualifying language (such as “enormous” and “interesting”) than the former. The second claim is a bit more tenuous because, overall, novels by white authors did not contain on average more adverbs and adjectives than novels by black authors, but some subsets of novels by white authors, such as prizewinning novels, did contain more modifiers than novels by black authors. In any case, we have a signal as to semantic and syntactic distinctions. And we find that narrative features are not significant. (64)
Table 2 from Richard Jean So and Edwin Roland, “Race and Distant Reading” (2020) showing the most distinctive language features of white and black authors in their study corpus of American novelists.
The variance in how novels by black authors use black features is about the same as the variance in how novels by white authors use white features. However, there is a significant difference in how novels by white and black authors use each other’s features. Specifically, the variance in how novels by black authors use white features is forty percent greater than the variance in how novels by white authors use black features. (65)
Now, what is striking is not only the magnitude and stability of the distinction between novels by white and black writers but also the fact that a number of texts appear where they should not be according to their binary labels…. Typically, computer scientists think of misclassifications as simple errors—like an algorithm’s misidentification of spam as a real e-mail. But when we begin with the belief that our categories may very well be tenuous, and we are interested in testing that possibility, misclassified texts provide information about what makes those categories unstable. (66-67)
… we found that novels by black authors are far more likely (nearly five times more likely) than novels by white authors to be misclassified. This means that if the distinction between white and black is tenuous, this indeterminacy is animated more by our corpus of novels written by black authors. (67)
Richard Jean So and Edwin Roland, “Race and Distant Reading” (2020) — Figure 1, p. 67.
How many features would we have to change in order to get the machine to predict [James Baldwin’s] Giovanni’s Room was written by a black author? We found that six out of the model’s fifty-seven features had to be altered for the machine to reclassify Baldwin’s novel as a novel written by a black author. These features, sorted by their contribution to the text’s misclassification from highest to lowest, are the words absolutely, very, course, appalled, might, and white. The first five features are characteristic of texts written by white authors and appear frequently in Giovanni’s Room, while white is predictive of black authorship but appears infrequently in the text. (69)
… what is most striking about the word white is how rapidly it disappears from the novel. By the end of the story, the word has completely vanished from the text. As we see in the above passage, white objects often appear in the text to prompt a set of social interactions, but the text quickly stops paying attention to those objects in order to pay more attention to the interactions. This tendency helps to explain the relative prominence of intensifiers in the text—they serve to elaborate and define those social interactions: absolutely, very, and of course. (70)
Indeed, the goal of this essay is to begin the hard work of developing a critical version of distant reading appropriate for the analysis of race and racial discourse in literature. We envision a reflexive method that is able to identify its own elisions while also pointing to new insights and opportunities for research.(72)
Richard Jean So, Redlining Culture: A Data History of Racial Inequality and Postwar Fiction (2020)
This is the main course website. There is also a course Canvas site for uploading assignments.
— (read only to end of p. 79, before the math begins)
Alan’s slides on relation between text-encoding and text-anaylsis (using Yin Liu’s table of paradigms of text in “Ways of Reading, Models for Text, and the Usefulness of Dead People”)
