A compositional morpheme tokenizer built on holographic reduced representations

Turns each word into one fixed-width vector by binding prefix $p$, root $m$, and suffix $s$ to unitary role vectors with circular convolution:

$$w = r_{\text{pre}} \circledast p \;+\; r_{\text{root}} \circledast m \;+\; r_{\text{suf}} \circledast s$$

Morphemes can then be algebraically recovered, and related words land near each other. It is an embedding layer for experimental models, not a drop-in replacement tokenizer.

@misc{tbc_morph_hrr,
  title = {A compositional morpheme tokenizer built on holographic reduced representations},
  author = {{TBC Research}},
  note = {F-009, catalogue version 1},
  howpublished = {\url{https://tbcresearch.org/research/morph-hrr/}}
}