Nov 12/18 NLP module, DSC 530 Very tentative syllabus (About 9 classes; & some quizzes; 2 - 4 assignments: English phrase structure & conversion between dependency & constituency; and between thematic roles and type composition; a left-corner chart parser; an oracle, and tree generator, for an NN shift-reduce constituency parser that could in principle be trained on WSJ (Wall Street Journal) corpus or Brown corpus; a POS tagger using word representations from TensorFlow, with a 5-word window (no HMM architecture).) 1. Linguistic structure -- morphology, phrase structure, features; -- unbounded dependencies; -- dependency structure as alternative 2. Phrase structure parsing (chart parsing); dependency parsing; CCG? Emph: left-corner chart parsing with top-down filtering Shift-reduce parsing; NN shift-reduce constituency parsing; 3. Semantic types and compositional interpretation; semantic parsing: phrase structure to surface logical form; Can use NN shift-reduce constituency parsing; relationship of surface logical form to FOL 4. Surface logical form to deep logical form: anaphora resolution, structural amendment and augmentation, operator scoping, deindexing, word sense diambiguation, canonicalization; relationship to FOL 5. Inference: FOL, natural logic; discourse inference; schema-based inference; taxonomic, temporal, and imagistic inference 6. HMMs & POS tagging; statistical parsing; NN-based parsing (LSTM -- e.g., Liu & Zhang, "In-Order Transition-based Constituent Parsing", TACL 2017, http://aclweb.org/anthology/Q17-1029; Dyer et al., "Recurrent Neural Network Grammars", NAACL-HLT 2016, https://www.aclweb.org/anthology/N16-1024; Dyer et al., "Transition-Based Dependency Parsing with Stack LSTM", ACL 2015. ) (Or possibly Cross & Huang, "Incremental parsing ... LSTM", 2016, https://arxiv.org/pdf/1606.06406.pdf; or Lewis/Lee/Zettlemoyer CCG parser; https://homes.cs.washington.edu/~lsz/papers/llz-naacl16.pdf; see as basis Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long Short-term Memory. Neural computation, 9(8):1735-1780.) [PROBABLY NOT: 7. Learning HMMs (Baum-Welch); learning PCF grammars (inside-outside)] [see http://www.cs.rochester.edu/~schubert/temp.html] [PROBABLY NOT: 8. Semantic role labeling; statistical semantic parsing] 9. The frontier: dialogue agents; knowledge & schema acquisition from text; Other topics: document retrieval (tf-idf relevance measure; LSA); document clustering; TensorFlow ML; CRFs for sequence processing; chatbots; QA, IBM's Watson & Wolfram|Alpha; MT; summarization; sentiment analysis; voice-based personal assistants