Riva An Intelligent Virtual Agent Reader

1
RivaAn Intelligent Virtual Agent Reader

• Center of Biomedical Engineering Bulgarian
  Academy of Sciences
• Technical Unuversity Sofia
• Technical University Sofia, city Plovdiv
• Plovdiv University Paisii Hilendarski.

2
(No Transcript)
3
Introduction IVA could be utilized for
Web-based interfaces, as personal computer
applications or other personal devices for
interactive television. IVA are an attractive
personage in games and amusements. But they could
find implementation and for serious tasks, for
instance as tools for psychological studies. Even
the current development of IVA assigns them a
place as in the research, as well as in the
industrial application.
The Main Objective of the project Riva The
project is inspired from the idea to develop an
intelligent virtual agent (IVA), with more
realistic behavior and sophisticated character.
The main objective of the project is to develop
an intelligent virtual agent capable of reading
natural language text on a certain topic,
assisting the process of students teaching and
reviewing movies. These capabilities are inferred
in the different meaning of the word reader,
thats why we named our agent Riva (R- from
reader, and iva from IVA ).
4
(No Transcript)
5

• The IVA Riva
• The agent will have a profession an assistant
  professor, and a hobby movie critique.
• Riva will have a visualized head and body. The
  lips movement will be synchronized with the
  syllable being pronounced. The eyes and head
  movement will be also controlled and
  synchronized, as well as the hands and body of
  the agent.
• Riva will learn by extracting facts and rules
  from natural language text files (popular movie
  reviews and lecture notes on Artificial
  Intelligence).

• She will present the lecturers notes with proper
  intonation, speed, mimics and gestures. Riva will
  also present and a practical lessons, will be
  able to ask students questions and will evaluate
  their answers.
• She will review movies (remember her hobby),
  outlining their pros and cons.
• Well lead an experiment in which Riva will
  evaluate the students work on an exam, assuming
  that
• The lecturers notes are the most authoritative
  source on the given topic.
• The students are supposed to reproduce the
  topic as closer as possible to the lecture,
  trying not to miss the key issues.

6
(No Transcript)
7

• The Experiment
• On this experiment, first the agent will learn
  the lecture, stored in a text file. She will
  convert it into distinct structural elements of
  the concepts, and will construct frame-concepts
  (a set of facts and rules, extracted from the
  lecture) for each recognized part of speech.
• On the next stage, Riva will receive the
  students works on the topic again in a text
  file. She will build the respective
  frame-elements in a way analogous to the
  described above. Each frame-element will be
  compared to the closer frame-concept. Thus, the
  similarities, differences and the contradictions
  will be found. Its assumed that as many the
  similarities are, as good the student write on
  the topic is.
• On the final stage, the agent will inform the
  teacher about the similarities, differences and
  the contradictions. The results will be discussed
  and a statistics concerning the most frequent
  contradictions will be stored (as for the current
  exam, as well as for the all exams held by the
  moment).
• In a similar manner, Riva will compare various
  reviews for a movie.
• The problem about finding implicit
  similarities/differences will be also discussed.

8
(No Transcript)
9

• Expected Results
• The development of an intelligent virtual agent
  reader, who has a visualized head and body and
  synchronized facial and body gestures.
• The development of a module for natural language
  (Bulgarian, in particular) speech pronunciation
  with proper intonation, speed and emotion. To
  achieve synchronization between lips movement and
  the words being pronounced.
• To develop a module for agent learning from
  natural language text files.
• To develop a module for building frame-concepts
  and for structuring the agents knowledge base.
• To implement the agent ability to present
  lectures and exercises with the proper speed,
  intonation, and gestures
• To implement the agent ability to evaluate
  students works (passed to the agent as a text
  files) from an exam.
• To implement the agent ability to compare
  different opinions on a given topic.
• To develop an Analyzing Synthesizing component
  that will lie on the base of composing movie
  reviews, recommending certain film, answering
  questions concerning a movie review already
  learned by the agent, comparing different reviews
  for a certain movie.

10
(No Transcript)
11

• A Further Work
• As a further development of this idea we plan to
  elaborate the agent character and behaviour,
  based on its knowledge and value system.
• As a last stage could be regarded the development
  of a virtual counterpart of a real human, which
  will help to verify the developed agent
  architecture and its value system.

Thank you for attention!