CSE467567 Computational Linguistics

1
CSE467/567Computational Linguistics

• Carl Alphonce
• cse-467-alphonce_at_cse.buffalo.edu
• Computer Science Engineering
• University at Buffalo

2
Todays lecture

• Syllabus and policies
• Announcement
• Overview of Computational Linguistics

3
Announcement

• Change of day/time for recitation,
• from Wednesday 400 PM 450 PM
• to Monday 600 PM 650 PM
• Let me know if you have a conflict with
• the new time.

4
What is Computational Linguistics?

• The study of techniques for processing natural
  human language by computer.
• computational techniques that process spoken and
  written language, as language Jurafsky
  Martin, pg. 2

5
Applications (extant envisioned)

• spelling checkers
• grammar checkers
• natural language interfaces
• conversational agents
• machine translation
• information extraction
• text summarization

6
General problem areas

• speech recognition
• natural language understanding
• information retrieval (finding sources)
• information extraction (extracting from sources)
• inference
• natural language generation
• speech synthesis

7
If only
8
Levels of processing

• phonetics/phonology sounds
• morphology word structure
• syntax sentence structure
• semantics meaning
• pragmatics goals of language use
• discourse utterances in context

9
Basic models

• state machines
• e.g. finite state automata and transducers
• formal rule systems
• e.g. regular and context-free grammars
• logic
• e.g. first-order logic, semantic networks
• probability theory

10
Basic algorithms

• state space search
• searching through possible hypotheses
• e.g. depth-first, best-first
• dynamic programming
• solving problems by combining solutions to
  subproblems without recomputation by storing
  subproblem solutions in a table

11
Ambiguity a pervasive problem

• An expression is ambiguous if there are two or
  more different meanings possible.
• Ambiguity exists at every level of linguistic
  representation
• E.g. I made her duck (pg. 4)
• I cooked waterfowl for her.
• I cooked waterfowl belonging to her.
• I created the (fake) duck she owns.
• I caused her to quickly lower her head or body.
• I waved my magic wand and turned her into
  undifferentiated waterfowl.

12
Short history

• Early language processing work in 1940s
• 1950s-60s saw split into symbolic and stochastic
  approaches
• formal language theory, parsing (top-down,
  bottom-up, dynamic programming)
• Bayesian approach used for OCR, authorship
  attribution

13
Short history (continued)

• 1970s produced several approaches to language
  processing
• stochastic
• logic-based
• natural language processing
• discourse modeling

14
Short history (continued)

• 1980s saw renewed interest in finite-state
  models (morphology) and probabilistic models
  (speech processing)
• 1990s joining of symbolic and stochastic
  approaches