Everything You Ever Said (EYES)

This is opportunity to play with Everything You Ever Said, an interactive work by Dr Nick Kelly (QUT, Australia) to go with an article in The Conversation (forthcoming).

Its purpose is to allow you to explore the meanings that machine learning models of the English language are making behind the scenes. EYES uses the Global Vectors for Word Representation (GLoVe) model to change the meaning of any piece of text (less than 2000 characters). It works using an understanding of English that gives every word a position within a 300 dimensional space, allowing us to perform calculations with words (such as addition or subtraction).

You can play with EYES by:

  1. Paste your text below
  2. Choose a "concept" (any one word) that you wish to "subtract" from your text (for example: fear)
  3. Choose a "concept" (any one word) that you wish to "add" to your text (for example: pleasure)
  4. Choose the "heat" for your transformation, where sliding to the right changes more words, and to the left fewer words (recommended to leave it where it defaults to start with)
  5. Type in your email. Results can take a while to generate so we need your email to send you a link. Emails are not stored after sending (I am an academic and this is a non-commercial project)
  6. Type "submit"

Song lyrics, poems, and speeches work particularly well. Try for example the Beatles' "Yesterday" with fear subtracted and enlightenment added, or put in your favourite speech and change it's core concept. Get inspiration from some examples.

Note: this version of EYES uses a stripped down model of English (6 billion tokens and 300 dimensions) based on limited processing power. The full model prefers the GLoVe Common Crawl model with 840 billion tokens and 300 dimensions