The use of ontologies in dialogue systems improves:

1
Introduction

• The use of ontologies in dialogue systems
  improves
• - The communication process
• - The engineering features
• The use of ontologies facilitates the management
  of the different types of knowledge involved in
  communication
• - Generic dialogue application rules
• - Application knowledge
• - Linguistic knowledge

2

• We present a dialogue system using ontologies
  for
• - Guiding the interaction
• - Processing the users interventions
• - Generating the systems messages
• The system supports textual input in different
  languages for different types of applications.
• The current system implementation has been
  applied to support dialogues in English, Spanish
  and Catalan for an application for collecting
  large objects from private homes.

3
Ontologies and the structure of the dialogue

• In the ontology, the concepts are described by a
  set of attributes or slots.
• Concepts are related to each other by the basic
  relations is_a and instance_of.
• All the information the application needs from
  the user is represented by a set of concepts
  (described by attributes).
• The dialogue system follows the Information State
  Update approach. The information states are
  obtained from the ontology.

4
Concepts may include peconditions that govern the
information that has to be asked to the user at
each state.
object_collection servicetype
information
cancellation
collection object address telephone Pco
attribute_value(object_collection, servicetype,
action)
5

• Ontologies describing domain concepts can also be
  included.
• Domain ontologies could be used to detect
  hyperonym and hyponym of an expected response.
• For example, in the Large Objects Collection
  application, the system needs information about
  the type of object the user wants to throw out.
• System What type of object you want to throw
  out?
• User An appliance
• The system could use an ontology describing the
  objects. If the concept appliance is classified
  in the objects ontology as an hyperonym of the
  information the application needs, the dialogue
  system would ask the user to be more specific.
• System What type of appliance
• User A refrigerator

6

• Domain ontologies can be used to avoid asking the
  user difficult questions which answer can be
  inferred.
• For example, the system needs information
  about whether the object the user wants to throw
  out is pollutant or not.
• The description of an object type in the object
  ontology could include the slot pollutant,
  indicating whether the object is pollutant or
  not. If the user answer is I want to throw out
  a refrigerator, the system can infer if it is
  pollutant or not from the ontology without asking
  it to the user.

7
Ontologies and the semantic processing of user
interventions

• Our system uses application-restricted grammars
  and lexicons. They are generated from the
  ontology representing the application knowledge.
• Application-restricted resources are efficient
  because they reduce ambiguity and simplify the
  interpretation process.
• The semantic interpretation is based on lambda
  calculus.
• Lambda calculus allows a simple and efficient
  interpretation process.

8

• The lexicon
• There are entries general to all applications.
• There are entries generated for the particular
  application.
• They correspond to concepts, attributes and
  values in the application ontology.
• The lexical entries consist of three fields
  string, category and semantic interpretation.
• The semantic interpretation associated with each
  lexical entry consists of a lambda function or a
  lambda value.
• Categories are augmented with syntactic and
  semantic features.
• The entry representing the verb to throw out,
  associated with the concept collection.

String to throw out Category
vcon(syn(tense(inf)),sem(con(collection)))
Semantic information ((l,X),collection,object,X)

9

• The grammar
• It is a definite-clause grammar.
• Semantic information is associated with each rule
  to indicate the order of interpretation of its
  constituents. It consists of a list of the
  numbers representing the constituents.
• Preconditions can be incorporated into the
  grammar rules to dynamically adapt the linguistic
  resources to the application requirements.
• The grammar rule for expressing the sentence to
  throw out a chair

s -gt vcon(syn(tense(T)),sem(con(C)))
indefngattr(syn(num(N)),sem(con(C))) (1 2)
10

• The parser
• It is a left-corner parser.
• It performs the syntactic and semantic analyses
  in parallel.
• It is implemented in Prolog.
• Once all constituents in the rule have been
  recognized, they are analyzed semantically
  following the order indicated in the semantic
  list associated with the rule.
• The semantic analysis consists of applying the
  lambda functions over the lambda values following
  the order indicated in the rule.
• The semantic interpretation process returns a
  list of words representing operations and their
  parameters (concepts, attributes and values).

11
Ontologies and the systems interventions

• The systems messages are generated from the
  attributes describing the application concepts in
  the ontology.
• The systems interventions consists of sentences
  asking or giving the attribute values.
• The attributes describing concepts are
  classifyied according to a semantico-syntactic
  taxonomy of attributes
• Each class is related to the linguistic
  structures expressing the consulting and filling
  of the attributes in the class

12
The basic attribute taxonomy

• Participants who_does, who_object, what_object
• Being is
• Possession has
• Descriptions and relationships between two or
  more objects of
• Related processes does

13
Conclusions and future work

• We propose the use of ontologies representing the
  application knowledge for improving the
  communication process and engineering features in
  dialogue systems.
• We have developed a system supporting textual
  input in different languages through the web.
• We are currently adapting the system for
  accepting voice input (using VoiceXML).

14

• We are currently working in facilitating the
  process of adapting the system to new
  applications. It includes
• - Defining new metaconcepts in the ontology
  representing the application knowledge.
• - Improving the process of obtaining
  semi-automatically the grammar and lexicon from
  the application ontology.
• Future work will also include providing virtual
  assistant about the contents of a particular web
  site.

15
References

• J. Bateman, B. Magnini and F. Rinaldi.The
  Generalized Italian,German,English Upper Model.
  ECAI Workshop, 1994.
• M. Gatius and H. Rodríguez. Adapting general
  linguistic knowledge to applications in order to
  obtain friendly and efficient NL interfaces.
  VEXTAL Conference, 1999.
• D. Milward and M. Beveridge. Ontologies and the
  structure of dialogue. Eigth Workshop on the
  Semantics and Pragmatics of Dialogue Catalog,
  2004.