Constrained Structure of Ancient Chinese Poetry Facilitates Speech Content Grouping
martyb writes:
[An] international research team theorized that the constrained structure of poetry serves as a mental template that allows readers and listeners to group creative poetic language into coherent meanings.
In order to test their hypothesis, the team focused on a genre of ancient Chinese poetry called Jueju, which has a highly constrained style. They generated artificial poems using a recurrent neural network so they could present novel Jueju poems to their participants, while controlling the poetic content.
Nearly eighty thousand ancient poems written over the course of five Chinese dynasties were fed into the neural network model, which then learned to create artificial poems based on the Jueju form.
The researchers synthesized each poem into a speech stream, removing the pauses, intonation, and other prosodic cues that a human speaker would produce, so that listeners had to rely on their knowledge of poetic structures in order to parse the stream.
Native Chinese speakers then listened to the artificial speech streams in an MEG[*] scanner, while the researchers aimed to detect neural signatures in the participants' brains that corresponded to the poetic structure. And indeed, the scientists discovered a brain rhythm of around 0.67 Hertz corresponding to the line structure of Jueju.
Even though the modern Chinese listeners were hearing each "pseudo ancient" poem for the first time and could not fully understand every phrase in the poems, they recognized the highly constrained structure and then actively grouped the poetic speech stream into lines according to their prior knowledge of Jueju. When the participants listened to the same poem for the second time, their brains had learned the structure, which allowed them to predict the forthcoming lines.
Read more of this story at SoylentNews.