What’s in a Conversation? (2022)

Conversation as a dataset, and the creation of paralinguistic hypertext to represent higher-dimensional information

Megan Prakash, with contributions from Wonki Kang

A conversation is more than just the words spoken. It includes common ground, conceptual topics, sentiments, hesitations, social cues, and many other higher dimensional features.

We used the YouTube video “Do You Feel American?: Immigrant Parents vs 1st Generation” as our dataset; it’s a conversation between children of immigrants and immigrant parents.

Our visualization encodes two sets of features:

(1) Shared concepts between the two groups, such as “impact of family ties” and “closeness to heritage.” Our intention is to represent the complex sentiments such that users can make meaningful comparisons.

(2) Hesitation and emphasis as the dialogue is spoken. We want to experiment with the expressive affordances of text to represent these paralinguistic traits.