Adjusting Interpretable Dimensions in Embedding Space with Human Judgments

arXiv:2404.02619v1 Announce Type: new
Abstract: Embedding spaces contain interpretable dimensions indicating gender, formality in style, or even object properties. This has been observed multiple times. Such interpretable dimensions are becoming valuable tools in different areas of study, from social science to neuroscience. The standard way to compute these dimensions uses contrasting seed words and computes difference vectors over them. This is simple but does not always work well. We combine seed-based vectors with guidance from human ratings of where words …

abstract adjusting arxiv compute cs.cl dimensions embedding gender human multiple neuroscience object science social social science space spaces standard study style tools type