Home // International Journal On Advances in Life Sciences, volume 16, numbers 3 and 4, 2024 // View article

Connotation and 3D Modeling from Limited, Raw Textual Descriptions

Authors:
Ella Berman
Mahiro Noda
Kailee Shermak
Zi Ye
David Rothfusz
Jiayi Chen
Thammik Leungpathomaram
Shuta Shibue
Chenxing Liu
Fernanda Eliott

Keywords: abstraction; connotation; memes; 3D-modeling; textual descriptions

Abstract:
In emotion-rich contexts, how do you comprehend the meaning behind your perception? This exploratory multiphase project seeks to gather insights into how abstraction and emotions travel different spaces. The investigated spaces are: images, human-made textual descriptions, mental models, and 3D (three-dimensional) scenes. In previous work, we described our project idea; here, we detail our pilot for Project Phases 1 and 2, in which a team first creates raw descriptions of memes (in addition to creating detailed descriptions and the Observer- Centered Dataset Attributes) so that the Phase 2 team, so-called modelers, read the raw descriptions and build a 3D scene as accurately and faithfully as possible to the meaning behind their perception of the description. Raw descriptions are created by “unsaying” (i.e., by identifying and removing the unsaid elements from a detailed description); and therefore, are more vague than alt-text since raw description purposefully leave details out (to see if the modelers “got” the message in spite of gaps). We designed a diagram to illustrate how modelers decided a 3D scene was complete, called “Accuracy and Faithfulness Gateways Diagram”, detailed here. We launched this project as a pilot to inform our methods to ensure objectivity and replicability. A key challenge in identifying the unsaid elements comes from making the implicit explicit, and our approach to accomplishing that can inspire frameworks for detecting biases and microaggressions in visual content and help to create cultural sensitivity awareness. We pinpoint our work’s social impact applications, which will be detailed in future work. Finally, investigating abstraction within and across spaces is notably relevant right now. In fact, as more people interact with generative AI platforms (such as AutoGen or Vertex AI), prompt designers deal with and add abstraction into a prompt as they instruct an AI-powered model to behave in certain ways.

Pages: 122 to 145

Copyright: Copyright (c) to authors, 2024. Used with permission.

Publication date: December 30, 2024

Published in: journal

ISSN: 1942-2660