1 / 16

 A LAD for OT

 A LAD for OT. Markedness in acquisition: Hypothesis: Universal markedness principles are genetically encoded, learning is search among UG-permitted grammars. Question: Is this even conceivable ? Collaborators: Melanie Soderstrom Donald Mathis Oren Schwartz. V. C Ons. C Cod. ‘1’.

jake
Télécharger la présentation

 A LAD for OT

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1.  A LAD for OT Markedness in acquisition: • Hypothesis: Universal markedness principles are genetically encoded, learning is search among UG-permitted grammars. • Question:Is this even conceivable? • Collaborators: Melanie Soderstrom Donald Mathis Oren Schwartz University of Amsterdam

  2. V COns CCod ‘1’ ‘2’ CVNet Architecture /C1 C2/[C1 V C2] / C1 C2 / [ COns1 V CCod2 ] University of Amsterdam

  3. s1 Local: fixed, gene-tically determined Content of constraint 1 Global: variable during learning Strength of constraint 1  Connection substructure Network weight: Network input: ι = WΨ a s2 i 2 1 University of Amsterdam

  4. V COns CCod PARSE (MAX) University of Amsterdam

  5. V 1 COns CCod 1 1 NOCODA University of Amsterdam

  6. CVNet Dynamics • Boltzmann machine/Harmony network • Learning: modification of Boltzmann machine algorithm to new architecture • Algorithm minimizes distance to correct output distribution • Stochastic activation algorithm: during processing, higher Harmony  more probable • Final state: local maximum guaranteed; if slow enough, global maximum probable University of Amsterdam

  7. Learning Behavior • CVNet can only learn grammars consisting exactly of the CV-theory constraints Con with differential strengths • No guarantee of strict domination • A simplified system can be solved analytically • Learning algorithm turns out to ≈ Dsi() = e [# violations of constrainti P] University of Amsterdam

  8. To be encoded • How many different kinds of units are there? • What information is necessary (from the source unit’s point of view) to identify the location of a target unit, and the strength of the connection with it? • How are constraints initially specified? • How are they maintained through the learning process? University of Amsterdam

  9. Unit types • Input units C V • Output units COns V CCod • Correspondence units C V • 7 distinct unit types • Each represented in a distinct sub-region of the abstract genome • ‘Help ourselves’ to implicit machinery to spell out these sub-regions as distinct cell types, located in grid as illustrated University of Amsterdam

  10. COns V CCod ‘N’ ‘E’ ‘back’ Connectivity geometry Assume 3-d grid geometry University of Amsterdam

  11. V COns CCod 1 1 1 1 1 1 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 Constraint: PARSE • Input units grow south and connect • Output units grow east and connect • Correspondence units grow north & west and connect with input & output units. University of Amsterdam

  12. V COns CCod 1 1 1 1 1 1 1 3 3 3 3 3 3 3 3 +1 1 3 3 3 3 3 3 3 3 1 +1 3 3 3 3 3 3 3 3 Constraint: ONSET • Short connections grow north from V units and connect to COns unit University of Amsterdam

  13. Direction of projection growth • Topographic organizations widely attested throughout neural structures • Activity-dependent growth a possible alternative • Orientation information (axes) • Chemical gradients during development • Cell age a possible alternative University of Amsterdam

  14. Projection parameters • Direction • Extent • Local • Non-local • Target unit type • Strength of connections encoded separately University of Amsterdam

  15. Connectivity Genome • Contributions from ONSET and PARSE: • Key: University of Amsterdam

  16. Learning Behavior • Simplified system can be solved analytically • Learning algorithm turns out to ≈ Dsi() = e [# violations of constrainti P] University of Amsterdam

More Related