Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore an innovative unsupervised algorithm that discovers hierarchical, context-sensitive structures in raw symbolic sequential data through this research lecture by Shimon Edelman from Cornell University. Learn how this computational approach processes both artificial stochastic context-free grammar data and real natural-language corpora, including raw transcribed child-directed speech, without requiring pre-labeled training data. Discover the algorithm's methodology for identifying candidate structures as patterns of partially aligned symbol sequences, accompanied by equivalence classes of symbols in complementary distribution within their contextual patterns. Understand how pattern significance is estimated using context-sensitive probabilistic criteria defined through local flow quantities in graphs where vertices represent lexicon entries and paths correspond to corpus sentences. Examine how new patterns and equivalence classes can incorporate previously identified structures, leading to the emergence of recursively structured units that enable highly productive and safe generalization by opening context-dependent paths absent from the original corpus. Gain insights into this groundbreaking demonstration of an unsupervised algorithm's capability to learn complex, grammar-like linguistic representations that are demonstrably productive, exhibit structure-dependent syntactic phenomena, and achieve strong performance on standard language proficiency assessments.
