The back-and-forth chatter this week has included thoughts on 1) how minds are depicted in fiction and whether those depictions are accurate; 2) how can linguistics inform how we conduct our poem iteration process; and 3) whether there is a locus for creativity when considering the way bird brains operate when producing song.
Ben’s already written in his post about Strange Bodies by Marcel Theroux (which I’ve not read yet), and “Shakespeare’s Memory” by Jorge Luis Borges (which I luckily have a copy of, and promptly read). Shakespeare’s memory, which is bequeathed to the narrator in a strange verbal agreement, emerges in the narrator’s mind in disconnected fragments and sensations triggered by random associations with things in the narrator’s world. Many of the memories are aural in nature, which I find pleasing because it relates directly to our experimental birdsong poem project while also ringing true to Shakespeare’s own marvelous ear, and his ability to transcribe everything from romance to politicking in metrical verse.
Ben’s also had a conversation with a linguist this past week to get a sense of how the rules we’ve set for our birdsong poem project could be improved. One of the points the linguist raised was if we’re manipulating our poem pattern with random mutations, we need to consider what the minimum unit of mutation is. Consider these lines from Shakespeare’s Sonnet 73*:
That time of year thou mayst in me behold,
When yellow leaves, or none, or few, do hang
Upon those boughs which shake against the cold,
Bare ruined choirs, where late the sweet birds sang;
Is the unit a word (“against”), or a syllable (“be” or “hold”), or a subclause “where late the sweet birds sang”?
For now, we’ve been mutating our poem at the word level, but for each iteration, the mutation is either completely random (“birds” replaced by “swam”) or grammatically consistent (“birds,” a plural noun, replaced by “socks”).
While we continue following these rules, one of my tasks this week is to set up rules to have Ben manipulate a new poem or piece of text on a syllabic level. This could either mean random syllable replacement (“behold” becomes “nahold”—something nonsensical) or associative syllable replacement (“behold” becomes “rehold”—which means something else).
Why are we doing this, you ask? Because we’re both curious what randomness leads to in terms of meaning-making. Ben put it best: If we try to faithfully copy the zebrafinch juvenile’s learning process, we’ll end up reproducing the poem I first wrote—instead, we’re using the model of the finch’s learning system to lead us to someplace unexpected and potentially exciting. I find this to be a satisfyingly artful ambition in our very scientifically structured process of poem writing.
Meanwhile, learning about zebrafinch mimicry has got me thinking about birds that create original songs or improvise wildly. I’m curious about the models we have for understanding how improvisation or “creativity” occurs in other bird species. We talked a little about this and I have some new reading material on the evolution of vocal learning in birds, but I’ll leave that train of thought for later blog post.
*No, I do not know the Sonnets by heart, I just Googled, “Shakespeare sonnet bird.”
Geetha and I continue to work on our poem algorithm. I am excited to see how it is going, but at the moment she is writing and I am merely supplying random words without any knowledge of the content that she is creating. As this poem continues, we have developed a few new ideas for modifying the algorithm. Geetha will design the next set of rules and I will be the writing agent. In the meantime, I have been looking into alternative mechanisms for generating noise in the writing process. I met with Charles Chang, a linguistics professor, to explain the project and get some inspiration from models of human speech.
To capture the bird’s process of injecting acoustic noise into the song, I’ve been sampling words at random (using a uniform distribution) from the dictionary. However, computer algorithms that generate poems that humans can read and enjoy select words based on the transition probabilities between words that are observed in the English text. It would be interesting to incorporate this type of approach into our algorithm. Incorporating transition probabilities to bias the randomness should allow the noise to produce something more coherent. Another cool feature of this approach is that it could be extended from words to syllables. This would allow the generation of new words similar to English words, but that do not currently exist. Hmm…