Assignment 8

After reading The Subtext of a Black Corpus, it was interesting how we could use machine learning to uncover new meaning and ideas by using literature that means something to us. I never realized how machine learning could be potentially used to empower and represent social groups through mere text. Machine learning could be used to detect certain keywords or the arrangement of words that are unique in texts with themes related to identity, and if we could train multiple texts with similar themes, I’m excited what kind of narrative would result from it. Right now, machine learning only understands the standard language, but if it could understand different dialects, I think it could help reveal underlying messages within specific texts to those who are not familiar with the works. The current machine learning is mostly built from standard (white) language, so we could diversify the system by training texts written by people of color who do not speak the standard language.

Ross Goodwin’s training for image captioning model is fascinating because it can be used to identify certain behaviors and tendencies of people when they write captions. It was interesting to see how he trained the model with poetry text to generate something original. Although the result isn’t perfect and wouldn’t make much sense most of the time, I think it’s an interesting concept to explore for my final project.

The readings inspired me to pursue using a text dataset that explores the gender breakdown of Time Magazine covers where data was collected from participants who were asked to classify the gender of the person in the magazine cover since 1920s.