Execution of Imperative Natural Language Requisitions based on UNL Interlingua and Software Componen - PowerPoint PPT Presentation

Loading...

PPT – Execution of Imperative Natural Language Requisitions based on UNL Interlingua and Software Componen PowerPoint presentation | free to view - id: 1141c9-ZTUwN



Loading


The Adobe Flash plugin is needed to view this content

Get the plugin now

View by Category
About This Presentation
Title:

Execution of Imperative Natural Language Requisitions based on UNL Interlingua and Software Componen

Description:

... concepts of the application domain and are shared with the Domain Ontology. ... SeMaComp searches the Component Ontology to find methods related to the action ' ... – PowerPoint PPT presentation

Number of Views:81
Avg rating:3.0/5.0
Slides: 27
Provided by: flavial2
Category:

less

Write a Comment
User Comments (0)
Transcript and Presenter's Notes

Title: Execution of Imperative Natural Language Requisitions based on UNL Interlingua and Software Componen


1
Execution of Imperative Natural Language
Requisitions based on UNL Interlingua and
Software Components
  • Flávia Linhalis and Dilvan de Abreu Moreira
  • flavia, dilvan_at_icmc.usp.br
  • University of São Paulo Brazil (ICMC-USP)

2
Outline
  • Introduction
  • Related Works
  • The UNL Project
  • The HERMETO System
  • The Proposed System
  • The Component Ontology
  • Web Course Management Example
  • Conclusions and Future Works

3
Introduction
  • The Idea of Natural Language (NL) interfaces is
    appealing
  • Convenient
  • Restricted context
  • Several works have pursued the idea of accessing
    functionality using NL interfaces.
  • Our goal
  • execute user requisitions described in several
    restricted NL.
  • Use UNL ? Interlingua

4
Introduction
  • User requisitions refer to
  • Sentences with high level of semantic abstraction
  • Specific domain
  • Example
  • Add student Flávia Linhalis to the Hypermedia
    couse.

5
Introduction
  • To achieve our goal retrieve and activate
    software components to execute NL user
    requisitions. We have to
  • Convert requisition to UNL (Universal Networking
    Language) sentence.
  • Extract semantic relevant information from the
    UNL sentence ? to retrieve appropriated software
    components.
  • Advantage UNL is an interlingua.
  • Requisitions, expressed in different human
    languages, can be translated into the same UNL
    representation.

6
Introduction
  • NL ? UNL conversion HERMETO.
  • SeMaComp (Semantic Mapping between UNL relations
    and Components)
  • Analyses the UNL representation and identifies
    necessary components, methods and arguments.

7
Related Works
  • First efforts to execute user requisitions
    expressed in NL ? later 70s.
  • NLC ? user can type English commands and watch
    them executed on the screen.
  • Matrices and tables domain
  • NaturalJava ? same idea. Generates Java source
    code.
  • Both systems are very limited ? input in a
    restricted algorithmic fashion.

8
Related Works
  • OAA (Open Agent Architecture) and SOTA ?
    components and agents get a higher level of
    abstraction.
  • Our differential and advantage
  • Natural language requisitions are first converted
    to an interlingua (UNL).
  • Several languages can be used to describe users
    requisitions.

9
The UNL Project
  • Interlingua ? expressive power to represent
    relevant information of NLs.
  • For each NL two systems should be developed
  • "Enconverter" ? converts natural language into
    UNL.
  • "Deconverter" ? translates texts from UNL to a
    natural language.

10
The UNL Project
  • UNL represents sentences using three elements
  • Universal Words (UWs) Each UW relates to a
    concept represented as an English word.
  • UWs can be optionally supplied with semantic
    information to restrict its meaning.
  • Examples
  • book
  • book(iclgtpublication)
  • book(iclgtreserve)

11
The UNL Project
  • Relation Labels (RLs) express semantic relations
    between UWs.
  • RLs are represented as a pair relation_label(UW1,
    UW2).
  • Example
  • obj (move, table) defines a thing that is
    affected by an event. The example means the
    table moved.
  • Atribute Labels (ALs) express additional
    information about UWs (verb tense, intention,
    emphasis, etc).
  • Example obj(eat._at_past, apple._at_pl).

12
The HERMETO System
  • It is a UNL Enconverter, similar to the UNL
    Center Enconverter program, but HERMETO is more
    user-friendly.
  • It can convert several natural languages into
    UNL.
  • It receives as input a dictionary and a grammar
    for each target language (It works with one
    language at a time).

13
The Proposed System
  • HERMETO ? dictionary and grammar to convert
    imperative NL sentences into UNL.
  • UNL sentence is used as input for the SeMaComp
    (Semantic Mapping between UNL Relations and
    Components) module
  • It identifies what components, methods and
    arguments will be required to execute the UNL
    requisition.
  • It uses the Component Ontology, that shares some
    concepts with the Domain Ontology.
  • Components can access the Domain Ontology (using
    the Protégé API).

14
The Proposed System
  • Component Ontology

15
The Proposed System
  • Component Ontology
  • It is instantiated according to the application
    domain components.
  • Instances of class OntoDomainConcepts
  • Correspond to concepts of the application domain
    and are shared with the Domain Ontology.
  • Instances of class Component
  • Correspond to components of the application
    domain that can be related to one or more
    concepts of the Domain Ontology.
  • Example class Teacher (Domain Ontology) could be
    associated to Component TeacherComponent
    (instance of class Component).

16
The Proposed System
  • Component Ontology
  • Instances of class Actions
  • imperatives verbs related to the application
    domain.
  • Each action is related to one or more methods,
    and each method is related to one action.
  • Class UNLRelations
  • Relates UNL relations to information about the
    components.
  • Its goal is to indicate the mapping between a
    particular relation_label(UW1, UW2) and the
    components, methods, arguments and actions in the
    Component Ontology.

17
Web Course Management Example
  • Components created and related to concepts User,
    Student, Candidate, Course, Class, Monitor and
    Administrator.
  • Domain Ontology created.
  • Component Ontology instantiated according to the
    components.


18
Web Course Management Example
  • Special attention is given to class UNLRelations.
  • Identified UNL Relations
  • obj (UW1, UW2) ? UW1 is an action. UW2 is a
    concept (instance of Component class)
  • nam (UW1, UW2) ? UW1 is a concept and UW2 is an
    instance of this concept.
  • gol (UW1, UW2) ?UW1 is an action and UW2
    indicates the object that will be altered by this
    action.
  • mod (UW1, UW2) ? UW2 relates to what is going to
    be modified in UW1.

UNL-Component Mapping
19
Web Course Management Example
  • Dictionary and grammar rules created for the
    domain.
  • Examples of valid imperatives sentences to serve
    as input

- Create course Operating Systems. - Delete
course Java. - Add student John Smith to the
class xxx. - List classes of teacher Susan. -
Delete administrator Mary from the course Java. -
Update course Java candidate name from Mary Smith
to Maria Smith.
20
Web Course Management Example
  • NL requisition converted using HERMETO. Example
  • Delete administrator Mary from the course Java.
  • UNL relations will serve as input to the
    SeMaComp.

obj(delete,administrator) gol(delete,course) nam(a
dministrator,Mary) nam(course,Java)
21
Web Course Management Example
  • SeMaComp separates the tokens and classify them
    using the Component Ontology.
  • For the UNL requisition example, SeMaComp will
    identify the following relevant information.

Action delete Main Component
administrator Other Component course Argument
Mary Argument type administrator Argument
Java Argument type course Number of arguments
2 Return type none
22
Web Course Management Example
  • SeMaComp searches the Component Ontology ? to
    find methods related to the action delete and
    to the concept administrator.
  • Result deleteAdmin and deleteAdminCourse.
  • Using the Component Ontology, for each method, it
    identifies
  • number of arguments
  • argument types
  • return type
  • SeMaComp uses this information to find which
    method is the most suitable to execute the
    requisition.
  • The method is executed and, during execution, can
    query or modify the Domain Ontology.

23
Conclusions and Future Work
  • SeMaComp can be used in different application
    domains. It is necessary to
  • build the appropriate software component set.
  • define the dictionary and the grammar rules to
    HERMETO.
  • create instances of the Component Ontology.
  • define the Domain Ontology.
  • UNL is an interlingua ? several languages can be
    accepted to describe user requisitions.

24
Conclusions and Future Work
  • SeMaComp is limited to the information given by
    the user at the NL language requisition
  • extend the Component Ontology to support context
    information.
  • Perform semantic mapping using not only
    imperative sentence structures, but also
    interrogative and conditional sentence
    structures.
  • Provide a restricted NL interface that could be
    used by other systems.

25
Acknowledgments
  • Thanks to
  • Interinstitutional Center for Computational
    Linguistics (NILC - São Carlos/Brazil) for making
    the HERMETO system available to be used in our
    research.
  • University of São Paulo and Araraquara University
    Center (Uniara) for the financial support of this
    project.

26
Execution of Imperative Natural Language
Requisitions based on UNL Interlingua and
Software Components
  • Flávia Linhalis and Dilvan de Abreu Moreira
  • flavia, dilvan_at_icmc.usp.br
  • University of São Paulo Brazil (ICMC-USP)
About PowerShow.com