John Bryant

• PhD (2008) in Computer Science at U.C. Berkeley
• Researcher at International Computer Science Institute (ICSI)
• Email: jbryant at icsi dot berkeley dot edu

Research Interests:

• Systems for deep understanding of language
• Construction grammar and computational construction grammar
• Computational psycholinguistics
• Probabilistic models of grammar

Dissertation:

• Title: Best-Fit Constructional Analysis (download)
• The system uses a psychologically-plausible best-fit probabilistic construction-based model of parsing and interpretation (described in chapters 2, 4, 5) to:
  • Model and interpret complex linguistic phenomena like argument structure constructions, active/passive/control, and questions (chapter 6)
  • Model a reading time experiment from the psycholinguistic community (chapter 7)
  • Interpret parent-child interaction in Mandarin (which is a pro-drop language) (chapter 8)
  • We are in the process of extending the system for morphology, metaphor and mental spaces.
• To build a probabilistic model over feature structures, I designed a factored (best-fit) model over syntax and semantics that decomposed the likelihood of the feature structure using common independence assumptions into the product of a set of conditional probabilities. (Chapter 2 and 4)
• To parse the input utterance, I defined a (incremental) probabilistic left-corner parsing algorithm:
  • The parser's probability model was based on work by Manning and Carpenter and Stolcke. (Chapter 5)
  • Nonlocal semantic information was pushed along the parser stack(s), to compute the likelihood of an interpretation given the syntactic cues (e.g. the words and their order) seen so far. (Chapter 5)

Software:

• The analyzer/ontology software is written in java.
• Luca Gilardi built a wonderful gui for the analyzer.
• Download the best-fit analyzer + gui here.
• My old construction-based analyzer is here

Papers:

• Jerome Feldman, Ellen Dodge and John Bryant. A Neural Theory of Language and Embodied Construction Grammar. To appear in The Oxford Handbook of Linguistic Analysis. Edited by Bernd Heine and Heiko Narrog. (2009) (This is a draft version with barely legible figures)
• Eva Mok and John Bryant. A Best-Fit Approach to Productive Omission of Arguments. In proceedings of the Berkeley Linguistics Society. 2006.
• Matt Gedigian, John Bryant, Srini Narayanan and Branimir Ciric. Catching Metaphor. In proceedings of the workshop on Scalable Natural Language Understanding (ScaNaLU). 2006.
• Eva Mok, John Bryant, and Jerome Feldman. Scaling Understanding up to Mental Spaces. In proceedings of the workshop on Scalable Natural Language Understanding (ScaNaLU). 2004.
• A Brief Introduction to the Junction Tree Algorithm