Jake Williams, PhD

Drexel University

 

This talk will discuss how, when the naturally-harmonic distribution of social categories known as Zipf Law is applied in a semantic-mixture analysis to linguistic data, a dynamic Mixing Law (ML) emerges and takes the form of a high-pass filter. Our experimental work exhibits how mixtures of `well-formed' documents resolve 2 natural constants that depend on the physical properties of a corpus (collection of documents). The constants that naturally fit the ML to data are the mean wavelength of a latent, harmonic vocabulary of semantics (meanings) and the average number of unique symbols (word types) used to represent them in a given document. Analyzing the high-pass filter, we uncover an underlying stochastic process that can explain the ML via a generalization of Herbert Simon's preferential selection model. This harmonic `resonator' juxtaposes constant transmission power (word-type novelty) with the reinforcement of previously-transmitted harmonics (meanings). We propose these intuitive and seemingly necessary human communication requirements be viewed as robust, existential evidence of a latent-but-invariant semantic information space through which coherent, beam-like emissions (`well formed' documents) are exchanged and accreted between individual, growing, and fuzzily-aligned latent semantic vocabularies that we refer to as information bodies (us).  This is part of a continuing investigation that pursues a science which models and simulates the human psycho-linguistic function.