Word2complex

From titipi
Revision as of 21:33, 14 November 2021 by Varia (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

Word2complex

A workshop with Varia (Cristina Cochior and Manetta Berends)

This workshop was a play on word2vec, a model commonly used to create ‘word embeddings’. Word embeddings is a technique used to prepare texts for machine learning. After splitting the writing up in individual words, word2vec assigns a list of number to each individual word based on what other words they find themselves in the company of. Once trained, such a model deducts synonymous words from comparing contexts, or will suggest probable words to complete partial sentences. With word2complex Varia proposed a thought experiment to resist the flattening of meaning that is inherent in such a method, trying to think about ways to keep complexity in machinic readings of situated text materials.

Step 1: Cutting embeddings of words

Choose a body of texts that you would like to analyse. Count how many times words appear in this text. You can use a custom script or an on-line service. Pick one word that appears at least twice from the list.

Step 2: Embedding words

Use CTRL+F to find your word in the text that you are analysing. For each moment in which the word is used: describe briefly the context in which the word is.

Examples

Word: street (wordcount: 2)
Embedding 1: street -> activism

  • "We've been talking to people more involved in both intellectual and academic work on, for example, like Nadia on solidarity and Islamophobia, on thinking about colonial structures in organizing and activism on the street."

Embedding 2: street -> survival

  • "From Brussels, the food collection was allowed so I think in the streets you had long lines queueing up of people. They managed to do it in a way that was respectful of the social distancing measures basically."

Word: companies (wordcount: 2)
Embedding 1: companies -> crisis

  • "I think a magnitude level failure with our tax money, obviously, that went to their friends, but the collaboration around formulating and writing about extractivism, colonialism, settler colonialism, capitalism and how that manifests itself in moments of crisis like COVID, and the Shock Doctrine approach of companies and governments to implement things like track and trace and now probably also, what's it called certificate, of vaccine certificate."

Embedding 2: companies -> refusal

  • "Then there was an ongoing boycott by left-wing people of the companies that closed their doors to protest this."

Step 3: Identify/generate/complexify relations

Pick two words that have been embedded (this can include words that someone else embedded). Expand the semantic map below and feel free to adjust the connectors (they are starting points, not prompts)!

Example

shitstorm -> war on terror - "In the meantime, I got sidetracked by the shitstorm that has been happening around us for the last 10 years."

violence is to policies as shitstorm is to war on terror
violence is to a fascist street as shitstorm is to war on terror
shitstorm is to war on terror not as the state is to context

Semantic map

______ is to ______ as ______ is to ______
______ is to ______ not as ______ is to ______
______ is not to ______ as ______ is to ______
______ is to ______ as ______ is not to ______
______ ..... ______ ..... ______ ..... ______