Explore a comprehensive lecture on context-sensitivity and stochastic unification-based grammars delivered by Mark Johnson from Brown University in 2000 at the Center for Language & Speech Processing (CLSP), Johns Hopkins University. Delve into the challenges posed by non-local context-sensitive interactions in learning probabilistic grammars and understand the motivation behind using more general log-linear (MaxEnt) models. Discover the concept of pseudo-likelihood and examine experiments on learning stochastic unification-based grammars from corpora. Analyze the implications of this research and investigate the relationship between these models and optimality theory. This 1 hour and 26 minutes long talk provides valuable insights into advanced topics in computational linguistics and natural language processing.