Home Artifacts Historical costume descriptors bridge the gap between past and present | VTX

Historical costume descriptors bridge the gap between past and present | VTX


The next step is to send the detailed descriptions on index cards to Chreston Miller, a data and computing consultant at University Libraries, where he and his team of students oversee natural language processing, also known as machine learning. . There they compare the cards and predict which Costume Core terms should be used to describe the item.

“This project is quite interesting,” Miller said. “I’m a computer scientist, so when I tell people I work with fashion, they’re like, what connection is that? People in merchandising and fashion design have data they want to work with.”

Given the challenges of describing artifacts, especially those of a historical nature, Miller said one reason for having a controlled vocabulary is that the descriptions that accompany an artifact were written within a certain period of time, they therefore have certain phrases for different aspects of artifacts. “We read it today and we’re like, ‘What does that mean?'” Miller said.

Madhuvanti Muralikrishnan, 23, a graduate student and university library employee, is assisting Miller on the project while working on her master’s degree in computer science and applications. Its role is to implement the algorithms that associate a given description with the Costume Core vocabulary.

“In machine learning, we’re basically trying to teach the computer to do something that humans do, and it’s quite complex,” Muralikrishnan said. “We know what multicolor is, but how do you teach a program that and the intricacies involved? It was surprising to me. I read these things in class, and it’s very rewarding to be able to apply what I’ve learned and see the result.

Muralikrishnan said that after getting her graduate degree, she wanted to work in machine learning, and this project is directly related to her career goals. “I believe that computing is domain independent. So today we’re doing this for fashion. Tomorrow I can take the same algorithm and do it for medicine or whatever. Each domain has its own problems and challenges and it was fascinating.

So far, Miller has received over 5,000 cards with descriptions that need to be scanned and organized into a database so everything is accessible and searchable. The cards start out on paper, then Miller performs optical character recognition on the cards so the team can scan the card descriptions, a crucial step in making natural language processing possible.

Processing an item or note card can take up to 45 minutes. The team also uses an online database to share information. “You think, ‘OK, the color doesn’t seem too hard to describe,'” Miller said. “Well, how many different colored words can you use? Take blue for example. There is navy blue, baby blue, royal blue, etc. So you need to think about how you have a control vocabulary for this.

The terms are inconsistent over time. “I’ve noticed in a lot of our older recordings that some of the terminology isn’t quite accurate or is very specific to the period it was recorded in,” Smith-Glavina said. “One thing that struck me about the collection is the term bloomers. Bloomers was actually a term used in the 1850s to describe the earliest form of women’s trousers worn by dress reformers and by cyclists as a sports costume. But over time, that changed to balloon-like underwear, mostly for babies – the ruffled underwear they wear over their diapers that people tend to describe as panties puffy.

Virginia Tech’s rare fashion items are usually donated by people cleaning their parents’ or grandparents’ homes, and they find trunks containing many generations of old clothing.

“Most of the time they won’t know what the items are and it’s up to us experts to identify what the clothes are,” Smith-Glaviana said. “We’ll be asking donors questions like, ‘Who do you think wore this?’ And it will help us to determine if it is a woman or a man, an older or younger person, or even a child, because sometimes it is difficult to tell with historical clothing. We get a sense of the donor’s story, but usually their knowledge is quite limited.”

Their ultimate goal is to digitize the historical costume collection and make it accessible to anyone interested in researching historical costume. Accurate, understandable, and consistent terms allow the average website user or non-costumed history expert to search for website terms they understand even when the items they are looking for are very specific.

“For example, most people know corsets as corsets,” Smith-Glaviana said. “But they don’t realize that for much of history corsets are actually called ‘corsets’ and they are two different pieces of clothing. So, depending on the words they are looking for, they may never find what they are looking for.