In the previous activities, we worked on exploring biases in conversations with language models. The goal of this activity is to systematize that exploration by starting to build a structured dataset that measures biases in a smaller discursive unit: phrases (or sentences). To complete this asynchronous activity, you will first need to watch the following videos and then complete the two tasks.

Important: The deadline to complete this activity is Wednesday, August 21, at 11:59 PM.

Videos

Introduction Video: In this video, Luciana reminds us how this activity is connected to the course objectives.

EDIA Video: In this video, Luciana explains how to access EDIA and then explores four types of bias using EDIA: gender, age, death, and religion. She makes a total of six entries in the database from previous queries.

Task 1

The task for this activity is for each of you to make at least 15 entries in the EDIA database. If you want to make more, even better! These 15 entries must explore at least three different types of biases or controversial topics (either predefined or defined by you). As you use EDIA, the entries are saved automatically when you press the button.

Task 2

In this form, we ask you to reflect on the activity you carried out using the EDIA software. There’s no need to upload your entries for Task 2; they are automatically saved in the database as shown in the video. We will also begin thinking about the final project, which includes classroom practice.

Supplementary Material (Optional)

If you enjoy listening to podcasts, Luciana recently appeared in one with Jocelyn Dunstan, the professor from Chile who spoke in the first class, discussing some of these topics. It doesn’t just focus on these issues; other topics are also discussed, such as Luciana’s research work on natural language processing and images, as well as her role in scientific associations like NAACL and Khipu. This is optional, but if you’re interested, you can listen to the podcast at this link.