Illuminate

Illuminate scans a user-inputted body of text for adjectives that skew stereotypically male or female and offers users alternative, more gender-neutral phrases to use.

Comment

Inspiration

We were interested in gender-debiasing words in language. There are so many subtle biases built into our daily vocabulary that are difficult to detect. We felt that NLP can help detect these subtle biases.

What it does

The website is created for professional users wishing to gender debias text from recommendation letters, job performances and evaluations, etc. Essentially, the user can paste their text on the website and they are offered suggestion of various levels depending on the extent of the gender bias within adjectives they use. Some adjectives are flagged because they may be too derogatory or gendered against women. For others, the user is offered a suggestion for a gender-neutral synonym for the adjective.

How we built it

We used a combination of NLP tools like HuggingFace models including ones for token classification (parts of speech) and sentence similarity tasks, along with the OpenAI API for the word embeddings

Challenges we ran into

  • importing certain NLP libraries

Accomplishments that we're proud of

-conceptual idea

  • workflow design

What we learned

  • Using NLP libraries
  • Backend and Frontend development
  • Planning and executing a team project in a limited period of time as this was our first hackathon

What's next for Illuminate

Finding a way to mathematically improve our gender debiasing technique to detect subtler bias

Share this project:

Updates