Activity: Introduction to the Women Writers Word Vector Interface

By Cara Marta Messina

Description

This classroom-based activity will introduce students to the WWP’s Word Vector Interface and enable them to begin exploring larger thematic questions about the texts in Women Writers Online. The activity can serve as a starting point for incorporating WWO, word embedding models, and text analysis in the classroom.

This activity is broken down into several parts. First, students will be guided by the instructor through a series of searches and discussion questions. Using the first part as a framework, the second part will ask learners to work in groups and conduct some exploratory analysis with word vectors.

Instructors can then broaden the scope of this activity by asking groups to form a theory for how their explorations reflect some characteristics of the texts in WWO; students can then examine the texts in the WWO collection to refine their theories and consider how close and distinct reading approaches shape our understanding of early women’s writing.

Learning Goals

This activity will enable students to:

  • Gain a basic understanding of how word embedding models work.
  • Familiarize themselves with the WWO corpus and the Word Vector Interface.
  • Explore how the relationships between words in a corpus tie into larger cultural themes.
  • Reflect on the differences between close reading and text analysis at scale.

Activities

This classroom-based activity breaks down into two parts: the first part will be guided by the instructor based on a series of illustrative searches; using the first part as a framework, the second part will ask learners to begin making discoveries about the WWO corpus and to think about research projects they might pursue.

Introductory Exploration

Using the Word Vector Interface, show students these searches and ask them to consider the results:

  • On the home page, query “queen” and “king” using the “WWO Full Corpus” model and compare results.
  • Use the “Operations” function to query “queen” + “king” in the “WWO Full Corpus” and compare to the previous results.
  • Bonus: What happens when you search “king” and “queen” plus one of their top ten closest words? For example, consider the difference between “queen” + “kingdom” and “king” + “kingdom.”

Discovery Activity

Now, ask students to work in small groups and go to the “Clusters” tab. This function shows random clusters of words that are closely associated in that particular model’s vector space. We recommend continuing to use the “WWO Full Corpus,” for continuity with the first activity. Students can hit “Reset” to discover more clusters if none of the initial clusters are interesting to them. For more information on the difference between “Clusters” and other functions, we recommend going through this activity and explanation of “Clusters.”

Ask the groups to select a cluster and then begin using the different search functions to learn more about the relationship between words in this cluster. For example, one cluster may read “beauty, youth, virtue, delight, pure, etc.,” and a group might explore how the words associated with “beauty” change when they add or subtract other terms:

  • Basic query: “beauty”
  • Operations (Addition): “beauty” + “virtue”
  • Operations (Subtraction): “beauty” - “pure” or “beauty” - “virtue”

Ask students to consider the terms that result from their queries: which results are surprising? which match with what they were expecting to find? which seem especially interesting or difficult to explain? Ask them to develop an explanation for the associations that they are discovering between words and make a list of the terms that they think are particularly important for their theory.

Usage Investigation

Now that the groups have a list of terms and an idea about how they are associated with each other, they can begin investigating their use in the Women Writers Online collection to see how that theory might change. Students can click on each word to see search results in the WWO interface. Depending on the words under investigation, students might find that there are too many results for them to read each one; for example, the word “beauty” appears in almost 300 texts, so they may need to combine search terms or refine their results by genre or time period.

Next, ask students to consider whether the theory they developed using the Word Vector Interface is consistent with what they are seeing in the WWO texts. If not, can they develop a new explanation for what they have discovered in both the WWO interface and the Word Vector Interface? Finally, ask groups to share their theories, searches, and a few useful examples from WWO with the rest of the class.