Title: CAREER: Intelligent Generation of Text and Information Graphics
1CAREER Intelligent Generation of Text and
Information Graphics
- Motivation
- vital technical information involving scientific
or medical arguments may be difficult for lay
person to grasp - Proposal
- use AI to help technical experts produce
user-friendly arguments in text and/or graphics - use HCI methods to ensure effectiveness
- build demonstration system (GenIE) for genetic
counselors
This material is based upon work supported by
the National Science Foundation under Grant No.
0132821. Any opinions, findings, and conclusions
or recommendations expressed in this material are
those of the author and do not necessarily
reflect the views of the National Science
Foundation.
2The Genetic Counselor
- Meets with clients
- Informational/educational role
- Explanation of diagnosis of genetic condition
- Explanation of inheritance risks
- General explanation of background on genetics
- Counseling role
- Writes summary letter (1-2 pages) for client
3Client Issues
- Complex Subject
- probability and statistics
- hypothetical outcomes
- causality
- scientific and medical terminology
- diagrams may help
- Emotional Distress
- Readers ability to comprehend (innumeracy)
- Rapidly changing information
4GenIE Genetics Intelligent Editor
- Goals
- NOT to replace human counselor
- Reduce counselors effort
- artificial intelligence creates 1st draft of
letter - human counselor may revise or reject GenIEs
draft - Design online presentation client benefits
- supplementary graphics and animation
- links to other resources
- automatic updates
5Multiple Research Methods
Goal Evaluate presentation techniques under
controlled conditions in lab
HCI Experiments
Computational Model Building
GenIE
Goal concrete implementation of ideas for
demonstration and evaluation
Goal develop widely applicable computational
(AI) techniques for generating arguments
Corpus Analysis
Goal study corpus to understand how human
authors communicate technical arguments
6Research Methods HCI Experiments
- Before Formally evaluate effectiveness of
communication techniques before computational
models created, e.g. - How does layout of document affect comprehension
of arguments? - What types of information to present in text, in
graphics, or both? - Graphical depiction of argument structure
- After Evaluate communicative effectiveness of
presentations created by GenIE - ablation experiments to identify which factors
contribute or detract from communicative
effectiveness
7Research Methods Corpus Analysis
- Corpus Acquisition (text and graphics)
- genetic counseling summary letters, client
education documents (print and web) - Qualitative Analysis
- types of information graphic techniques
- analysis of argumentation (ex. predictive,
diagnostic, value-based, Toulmin-style,
dialectical) - Computational Linguistics Analysis
- develop coding scheme with intercoder reliability
- manually encode corpus
- manual and automated discovery of communication
techniques evolved by human authors
8Research Methods Computational Models
- Develop AI methods to
- represent the underlying scientific arguments and
reasoning of the experts - predict the audiences potential problems in
understanding, e.g., - complexity of causal explanation
- emotionally disturbing information
- reason about content (both text graphics),
organization, and layout to avoid predicted
problems - generate text and graphics based on above
9Analysis of Argumentation in Corpus
- Argumentation discourse that weighs evidence and
presents multiple points of view - An important dimension of argumentation in
letters in corpus diagnostic and predictive
reasoning - hearing loss was caused by mutation in gene
(GJB2) - if HD, then chance that others in family are
affected - Those parts of letter can be represented by
Bayesian (belief) network
10Bayesian Network
History/proband Age child
History/mother family history of deafness no
Genotype/mother one abnormal copy of gene GJB2
Genotype/father 2 abnormal copies of gene GJB2
50
50
Genotype/proband 2 abnormal copies of gene GJB2
Genotype/sibling 2 abnormal copies of gene
GJB2
Biochemistry/proband Connexin 26 abnormal
Physiology/probandnormal chemical equilibriumno
Symptom/proband deafness
Finding/proband facial defects no
Result/proband GJB2 test positive
Symptom/father deafness
Symptom/sibling deafness
11GenIE Project Summary
- Building demonstration system (GenIE) to help
genetic counselors write letters - Using HCI to ensure effectiveness of general
argument presentation techniques - Using AI to to model experts reasoning and
argumentation strategies - Techniques will be applicable in many domains to
problem of computer-assisted or automatic
multimedia generation of effective technical
arguments for lay audience