Ontolex: A Cognitive Model of Language Learning

1
Ontolex A Cognitive Model of Language Learning

• by Emily Fortuna
• NIST Information Technology Laboratory
• Rice University

2
Comprehensive searches for webpages
one-to-one method
Thesis topic

• You want all related topics
• Recipes
• Sam
• Dr. Seuss
• articles in other languages

3
Comprehensive searches for webpages
one-to-one method
Thesis topic
Dr. Seuss
Sam
Recipes
Exponential Growth
green eggs and ham
huevos verdes con jamón
le uova verde e prosciutto
oeufs et jambon verts
grüne Eier und Schinken
4
Create a Web of Related Concepts
Authors
5
Comprehensive searches for webpages
A better design Ontologies
Thesis topic
huevos verdes con jamón
green eggs and ham
oeufs et jambon verts
?? ?? ? ?
grüne Eier und Schinken
???? ????? ???? ???????
Zielone jajka i szynka
??????? ????? ? ???????
le uova verde e prosciutto
Virent Ova! Viret Perna!!
??????
p??s??a a??? ?a? ?aµp??
6
Ontology?

• A representation of how concepts in a domain
  relate to one another.
• In the form of
• A simple hierarchy
• A more complex web (can be recursive)

7
Ontology ? Hierarchy
flying monkey
Scarecrow
Slots give additional information For example,
fur color for Lion, Toto, and the flying
monkey.
8
Ontologies help make Computers more intelligent

• Drawing conclusions
• Example
• Glinda cast a spell.
• Dorthy cast a spell.
• Also
• Dorthys ruby slippers
• Totos ruby slippers

9
Ontology ? Web
Source Sowa, J. F. Knowledge Representation
Logical, Philosophical, and Computational
Foundations. Pacific Grove, CA Brooks Cole
Publishing Co. http//www.jfsowa.com/ontology/inde
x.htm, 2000.
10
Goals of Ontolex

• Create a second language acquisition computer
  model that is
• Easily understandable
• Python programming language
• Available
• Internet
• Reusable
• Ontologies and software design
• Any human language

11
Ideas powering Ontolex

• Ontologies
• Model for a model
• Sentence generation technique

12
Powering Ontolex ? One Unified Ontology
13
Not just Nouns

• All parts of speech can potentially derive from
  the concepts in Ontolexs ontology

Words derived
Concept
14
Ontolexs three-tier System
invent
Concept Layer (language independent)
15
Three-tier system Another perspective
The Ontology
Lexeme Language A
Lexeme Language B
Word Language A
Word Language B
16
Ideas powering Ontolex

• Ontologies
• Model for a model
• Sentence generation technique

17
Model design

• 3 sections
• Vocabulary building
• Form Sentences and Translation
• Unique Sentence Generation

18
Vocabulary Building

• Silent phase of language learning
• Ontology Builder
• Sentence Learner
• Database Viewer

19
Ontology Building Tree
20
Sentence Learner
21
Form Sentences and Translation

• Learner translates to native language to
  understand foreign language
• Learner uses sentence structures he/she has heard
  before
• Under development

22
Difficulty with Parse Trees

• Problem null subject languages, like Spanish and
  Japanese
• Nothing to put in NP
• Programming for individual transformations

23
Unique Sentence Generation

• Specifying Semantics
• Proposed model
• Concepts (ontology) ? Sentences
• User selects
• Verb
• Nouns (Subject, Object, etc.)
• Language for generated sentence

24
Semantics Example

• Example
• Verb eat
• Nouns human, bread
• Language Spanish

25
Sentence Generation Process

• Get Overall Sentence Structure
• Randomly select substructures
• Add lexemes
• Convert to words
• Receive user feedback

26
Sentence Generation

• Get Overall Sentence Structure
• Any combination of the any number of the
  following
• Subject
• Verb Phrase
• Object
• Indirect Object
• Example eat human bread ? O VP S
• Order acceptable for Spanish S VP O

27
2. Randomly select substructures
28
3. Create language tree with lexemes
man
eat
bread
29
4. Create language tree with words
man
eat
bread
The fat man eats bread.
30
5. Receive User Feedback

• Learns new syntactic and semantic combinations
• Uses reinforcement

31
Implications

• Short term
• Model
• Test language acquisition theories with computers
• Long term
• Ontologies ? more intelligent computers
• think and learn like humans
• Voice recognition
• Computer systems can interact with each other
• Better human-computer communication

32
Acknowledgements

• Thanks to
• Dr. Larry Reeker, my advisor
• NIST Information Technology Laboratory for
  funding this research and participating in the
  SURF program
• Rice University for informing me about this
  summer opportunity
• Dan Sandler, Michael Benza, Christopher
  Warrington, and the Python Tutor mailing list for
  programming language implementation discussion
  and suggestions in the early stages of this
  project.

33
http//ontolex.nist.gov
34
(No Transcript)
35
More Information

• In case you cant get enough

36
Website Layout
37
Model-View-Controller framework
38
Database Viewer
39
Form Sentences and Translation