Artificial Intelligence Expert Systems Natural Language Processing

1
Artificial Intelligence / Expert SystemsNatural
Language Processing

• Justin Gaudry
• July 5, 2007

2
Natural Language Processing

• Understanding
• Taking some spoken or typed sentence and working
  out what it means so that something can be done
  with it
• Generation
• Taking some formal representation and working out
  a way to express it in a natural language

3
Natural Language Systems

• Explore general theories of human language
  processing
• Do practical tasks
• Provide natural language interfaces
• Provide front ends to application systems
• E.g., Convert English sentence into formal
  database query

4
Understanding

• Spoken vs. typed
• Speech recognition
• Spoken much harder to understand than typed
• Input is raw speech signals taken from microphone
• Before analyzing, must determine what words were
  said

5
Generation

• Text planning
• Deciding what to say
• Start with a goal
• Determine how to say it
• Find sequence of words / phrases to achieve the
  goal
• Depend on flexibility involved
• For special-purpose application, can use template
  (like mail-merge)

6
Speech Synthesis

• For template, have human record relevant phrases
  and put together the phrases
• Have human record every word in the dictionary
  (and conjugations)
• Not viable
• Record phonemes and put together
• Phoneme basic distinctive units of speech sound
  by which morphemes, words, and sentences are
  represented
• Not intelligible
• Sounds dependent on those before and after
• No intonation, stress, or timing (rising in pitch
  when question, pausing)

7
Stages of Processing

• Phonology
• Morphology
• Syntax
• Semantics
• Pragmatics

8
Phonology

• Speech recognition by phoneme
• Raw speech signal is analyzed and sequence of
  words spoken obtained
• Benefits
• Flexible input
• Telephone
• High speeds
• Hands-free

9
Speech Recognition Process

• Convert analog signal to frequency spectogram
• Divide into high-pitch vs. low-pitch
• Match to phonemes (a in cat or sh in show)
• Small number of phonemes in any language allows
  for full library
• Match strings of phonemes to possible word from
  database

10
Problems

• Same word pronounced differently by different
  speakers
• Normal speech is fast may not be able to
  determine the end of one word and start of next
• Individual word recognizer vs. continuous speech
  recognizer
• Require simplifying assumptions
• Single speaker vs. speaker-independent
• Single speaker trains
• Restricted vocabulary yes and no, digits

11
Morphology

• Morpheme
• linguistic unit consisting of a word, such as
  man, or a word element, such as -ed in walked,
  that cannot be divided into smaller meaningful
  parts
• Useful for identifying meaningful components
• Parts of speech
• Present / past / passive tense
• Singular / plural

12
Syntax

• Grammar
• Rules of syntax making up the structure of the
  language
• Sentence ? subject predicate
• Subject ? noun phrase
• ? noun phrase
• conjunction noun phrase
• ? noun phrase
• prepositional phrase

13
Syntax

• Noun phrase ? noun
• ? determiner
• noun phrase
• ? adjective noun
• Predicate ? verb phrase
• ? verb phrase
• conjunction verb phrase
• ? verb phrase
• prepositional phrase

14
Backus-Naur Form Grammar

• Aka Backus normal form
• Terminal symbols
• Symbol or word in the language itself
• The, cat, dog (facts)
• Nonterminal symbols
• Map to other structures
• Noun phrase, prepositional phrase (rules)
• Start symbol
• Complete statement in the language
• Sentence
• Rewrite rules
• Rules defining grammar
• Sentence ? subject predicate

15
Parser

• Syntactic analysis
• Search for rules to match the sentence structure
• Derivation tree
• Depth first search through parse tree,
  backtracking when mismatch
• Chart parser
• Creates graph with edges for possibilities
• Once finds nonterminal phrase, doesnt need to
  revisit the words

16
Semantics

• Meaning of a sentence
• Compositional semantics
• Find meaning of nonterminal elements
• Put together to form entire meaning
• Useful for solving ambiguity problems

17
Pragmatics

• Apply semantic meaning to context
• John kisses Mary. He loves her.
• Kisses (John, Mary). Loves (He, her).
• Second needs the context of first to identify
  Loves (John, Mary) as well as the explicitly
  listed first paragraph.

18
Pragmatics

• Communication vs. action
• Communication
• Declare a fact. John loves Mary.
• Action
• Informing
• Requesting
• Promising

19
Pragmatic Action

• Do you have the time?
• Inform speaker if listener has the time
• Request that listener tells speaker the time
• Youre late.
• Inform listener of time problem
• Promise (category) irritation of the fact