Simulated distributional learning in deep Boltzmann machines leads to the emergence of discrete categories

Open Access
Authors
Publication date 2019
Book title Proceedings of the 19th International Congress of Phonetic Sciences, Melbourne, Australia 2019
Book subtitle ICPhS2019 : 5-9 August 2019, Melbourne Australia
ISBN (electronic)
  • 9780646800691
Event International Congress of Phonetic Sciences
Pages (from-to) 1520-1524
Number of pages 5
Publisher Canberra: Australasian Speech Science and Technology Association Inc
Organisations
  • Faculty of Humanities (FGw) - Amsterdam Institute for Humanities Research (AIHR) - Amsterdam Center for Language and Communication (ACLC)
Abstract There is a potential close correspondence between multi-level linguistic theories and bidirectional deep artificial neural networks. This paper shows that in a deep Boltzmann machine, simulated distributional learning of spectral content leads to the emergence of appropriate categorical behaviour, both along a one-dimensional continuum (three sibilant places) and along a two-dimensional continuum (five vowels).
Document type Conference contribution
Language English
Other links https://www.icphs2019.org/
Downloads
2019-icphs-Boersma (Final published version)
Permalink to this page
Back