Recent

• A probabilistic, language-agnostic syllabifier

• Do word frequencies in spontaneous speech form a power law distribution?
   

Current

• Do word frequencies in the spontaneous speech of children diagnosed with autism spectrum disorder form a power law distribution? 

• Classifying killer whale vocalizations.

• In conjunction with Dr. Mark Vandam, Department of Speech and Hearing Sciences, Washington state University

  • Analysis of child-directed speech in day-long recordings of children and their families. 

  • Modeling the speech of galactosemic children 

• Morpheein Enzyme Kinetics
  Burkholderia cenocepacia is an opportunistic pathogen particularly dangerous for cystic fibrosis patients.  It can cause a severe decline in lung function with the potential of developing into a life-threatening systemic infection.  Antibiotic resistance makes B. cenocepacia extremely difficult to treat.  The catalytic activity of enzyme 3-hydroxy-3-methylglutaryl coenzyme A reductase (HMG-CoA) provides an energy source for the pathogen.  Therefore, understanding HMG-CoA presents a promising opportunity to treat infection by B. cenocepacia.  Our initial goal is to develop mathematical models of the morpheein kinetics exhibited by Burkholderia cenocepacia.

• Orcinus orca Population Dynamics
  The orca or killer whale (Orcinus orca) is a toothed whale belonging to the oceanic dolphin family, of which it is the largest member.  Two distinct populations of Resident Killer Whales, known as the Northern and Southern Residents, occupy Canadian Pacific waters.  Northern Resident Killer Whales are listed as Threatened, and Southern Resident Killer Whales are listed as Endangered under the Species at Risk Act.  We seek to develop an agent-based model of the two Resident populations' dynamics to quantify the impact of food supply, toxic pollution, noisy waters, and physical disturbance that contribute to population wellbeing.

• ​Translation using a tree-structured interlingua
  SoTA translation models perform a mapping from one sequence of tokens to another using "transformers", multi-headed self-attention networks, divided into encoders and decoders.  While transformers are good at associating different meanings of a gloss (the string of letters representing a word) with different vectors, they cannot reliably associate the same vector with the same meaning in contexts that don't change the meaning.  For this and other reasons it is reasonable to conclude that transformers are approaching an accuracy ceiling, and human language is above that ceiling.  This research addresses translation with a novel technique that may function in the same complexity class as human language.  As a free by-product, this technique would also solve some open NLU problems.