Dependencies - PowerPoint PPT Presentation

1 / 12
About This Presentation
Title:

Dependencies

Description:

In addition to phonemes (basic sound units), prosodic features must be generated ... Predicative adjective (make) Infinitive (like) That-clause (think) Wh-clause (ask) ... – PowerPoint PPT presentation

Number of Views:40
Avg rating:3.0/5.0
Slides: 13
Provided by: VasileiosH9
Category:

less

Transcript and Presenter's Notes

Title: Dependencies


1
Dependencies
  • Vasileios Hatzivassiloglou
  • University of Texas at Dallas

2
Sample input and lattice
3
Sample random and default generation
4
Sample constrained MC generation
5
Text-to-speech
  • Another type of generation application, which
    also includes elements of translation
  • In addition to phonemes (basic sound units),
    prosodic features must be generated
  • Many applications
  • Automated telephone systems
  • Assist users with disabilities
  • Other innovative uses

6
Modeling dependencies
  • Local n-gram models cannot capture certain
    dependencies and additional information
  • Syntactic constraints
  • The velocity of the seismic waves rises to
  • Semantic constraints
  • Discourse constraints (sentence order and
    cohesion)

7
Syntactic constraints
  • Often represented with a context-free grammar
    that models phrase structure
  • Non-terminal nodes correspond to constituents,
    building blocks such as different types of
    phrases
  • Terminal nodes correspond to words
  • The grammar imposes a hierarchical, tree
    structure on sequences of words

8
Example grammar
  • S ? NP VP
  • NP ? DET NN DET NNS NP PP
  • VP ? V V NP
  • NN ? dog homework
  • V ? ate
  • DET ? the my
  • Note the recursive structure

9
Verb subcategorization
  • Additional constraints can be represented with a
    division into subcategories. We say that a verb
    subcategorizes for a particular complement
  • Object (eat)
  • Prepositional phrase (put)
  • Predicative adjective (make)
  • Infinitive (like)
  • That-clause (think)
  • Wh-clause (ask)

10
Probabilistic context-free grammars
  • Associate a probability with each production rule
  • All alternatives for a non-terminal must sum to
    one
  • Robust
  • Allow for comparison of alternative parses
  • Do not capture local dependencies

11
Semantic constraints
  • Start with predicates corresponding to verbs (and
    some nouns)
  • Each predicate has arguments that need to fulfill
    specific semantic roles
  • Example eat
  • Alternative Semantic grammars directly capture
    semantic distinctions by differentiating the
    non-terminals

12
Reading
  • Introduction to Section 3.2, Sections 3.2.1
    3.2.2
  • Introduction to Chapter 11, Section 11.1
Write a Comment
User Comments (0)
About PowerShow.com