CPSC 477 Study Guide - Spring 2018, Comprehensive Midterm Notes - Verb, Parsing, Noun Phrase
12 Oct 2018
School
Department
Course
Professor
CPSC 477
MIDTERM EXAM
STUDY GUIDE
Fall 2018
Borrows from linguistics, AI, build systems
web.stanford.edu/~jurafsky/slp3
ntlk.org
Tuesday May 9th 2 PM final exam
Core: parsing, part of speech tagging, text generation, semantic analysis
Yale-lily.github.io
Matrix multiplication, baye, derivative of sigmoid function
Johns Hopkins Jason Eisner
Industry: Salesforce
tangra.cs.yale.edu/newaan
112
grounding – things being said are grounded to physical object/ object
vaccines _____ autism
how many different ways to say vaccines cause autism
- Fake news
- Facts
- Speculation, hedging
- Complex vocabulary and sentence structures
News articles easiest to parse vs fiction
Graity’s Rainbo
- Most famous for being book started but never finished
Poetry even worse
find more resources at oneclass.com
find more resources at oneclass.com
Finite state machines
Chomsky hierarchy
- Different grammatical formalisms
- More power to desrie setees ad strigs
Input -> noisy channel -> output, figure out input
First systems made at Bell Labs 1950s
Taxonomy of objects, operate using first order logic
Want to produce English sentence, produce how plausible sentence is grammatical
- M-gram modeling?
- Approximate probability of sentence
Natural language -> logical form -> matched against universe (nlu)
1994 – 2010
- Statistical modeling
- Read text in English, French -> which to which using probabilities
- Combine info
And then machine learning
Be able to evaluate performance of system (shared tasks)
- chatbots have well defined task
- how well they can answer
- statistical analysis on results
Why NLP hard
Time flies like an arrow
- Multiple interpretations (nouns + verb use)
Mary and Sue are mothers
- How many relations
- 2
ambiguity
- multiple meanings
- part of speech use
- pronunciation
more on ambiguity
- computer languages try to avoid by design
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
Borrows from linguistics, ai, build systems web. stanford. edu/~jurafsky/slp3 ntlk. org. Core: parsing, part of speech tagging, text generation, semantic analysis. 112 grounding things being said are grounded to physical object/ object vaccines _____ autism how many different ways to say vaccines cause autism. Most famous for being book started but never finished. More (cid:862)power(cid:863) to des(cid:272)ri(cid:271)e se(cid:374)te(cid:374)(cid:272)es a(cid:374)d stri(cid:374)gs. Input -> noisy channel -> output, figure out input. Taxonomy of objects, operate using first order logic. Want to produce english sentence, produce how plausible sentence is grammatical. Natural language -> logical form -> matched against universe (nlu) Read text in english, french -> which to which using probabilities. Be able to evaluate performance of system (shared tasks) Pronunciation more on ambiguity computer languages try to avoid by design: specific constructions, arithmetic: precedence, parenthesis, specify types of variables ppl tried to invent languages based on logic unambiguity. Constructed language noun-noun phrases or adj + noun + noun + prepositional phrases.
