CHALLENGES APPROACHES/SOLUTIONS
  
ONTOLOGY Challenges
Ontology Creation/Population 
Tasks <-> Tools - reusable across domains 
Understand a process model (and humans
role in this) 
Semantic Web 
User-centered process view 
Convert the (HCI) disbelievers and keep them
practicing 
"top" or core ontology (use this to bootstrap
new domains), Ontology integration 
Rapid customization (to specific domains) 
Use domain specific ontologies to organize
massive documents 
Find, learn, collaboration with domain
ontology creators 
Integration of shallow/deep methods
ONTOLOGY Problems
- Ontology quality
- Access to info, knowledge visualizations
- Understanding
- Ambiguity
ONTOLOGY Approaches
Relation of HLT to ontological tasks 
KR, linguisits, & ontologies to jointly
address 
Component based methods for 
Life cycle 
Re-use 
Decomposition 
Use HLT to support knowledge audits
> Identify IP -> innovation 
Context capture 
Controlled, language management
ONTOLOGY Solutions
Plug-in (for IE) 
Semantic Web 
Tools to leverage small ontologies ->
large ontologies
 
  
 
SUMMARIZATION Challenges: 
level/depth of analysis/representation (E.g., Speech acts,
RST, semantic rels) 
Sumarization presentation/visualization 
Speech (not good for long texts) 
Indicative vs. inforamtive, concepts vs. ideas 
Action-oriented summaries (e.g., executive/management 
summaries)
 
SUMMARIZATION Solutions
- Analysis -> transformation -> presentation
21 Feb 2007  21:45Generated by HTML_ToPDF at rustyparts.com - 1 -
Notes
  
MULTILINGUAL Problems
Relational between cultures, languages, lexical resources,
ontologies 
Domain knowledge 
Fine-grained linguistic knowledge (e.g., stylistic details) 
Size, complexity 200 languages -> 39k language pairs 
Language invisibility
large-scale, robust NLP 
Adaptation/integration of semantic resources 
Content-driven hypertextual authoring 
Cross-lingual news linking 
Advanced software technologies/platform 
Communication/transaction success
 
 
 
MULTILINGUAL Solutions
resources: wordnet, euronet, application
database, text resources 
Interlingua approach 
Statistical -> deeply annotated data + machine
learning 
Translation memories + ML 
Multimodal/multimedia sols 
Multiple ontologies tailored to users, tasks
  
MULTIMEDIA Challenges 
- Processing centralized/mobile
- Privacy, security, scaleability
Remembering + forgeting 
multilingual and multisource IE incremental
information building 
cross-document co-ref resolution
MULTIMEDIA Solutions
Location-based services 
"forgetting"
Input to a Technology Road Map:
Enabling Technologies/Infrastructure
Mobile communications 
Push service failures (e.g., pointcast) 
Satellite communication bankruptcy 
Fibre explosion
Services
21 Feb 2007  21:45Generated by HTML_ToPDF at rustyparts.com - 2 -
Notes
video on demand failure need for content based access
Resources
RDF, DAML, OIL? 
Ease of integration 
IE, NE
Fundamental/Hard Problems
Noisy Speech Recognition 
Non-literal language 
Semantic web (e.g., who is going to populate it)
Ontologies
Auto Web Taxonomy Generation 
High Quality MT
/\
||
- Tools for ontology generation, merging
Free CYC?
Summarization
"conceptual" or "content" level diff (email, documents, patents)
Query dependent, Multiple perspective Summarization (representation and output)
/\ /\ /\
|| || ||
entity discourse co-ref
Multilingual
interlingua 
deeply annotated data + ML 
user appropriate translations 
English Interlingua
Multimedia
21 Feb 2007  21:45Generated by HTML_ToPDF at rustyparts.com - 3 -
Notes
personalized content based news 
multimedia I/O (maps, gesture)
/\
||
multimedia data and annotation (images, maps, video, medical)
Standards
Process Reusable interchangeable modules (e.g., POS, NE) 
Data (XML, text encoding, W3C)
NLP
Robust, deep language processing (e.g. LFG parsing which is fast but inaccurate still)
KM/Information Integration
Integrated mining, query of mail, DB, process knowledge
CORE ENABLING RESOURCES
- (intelligent) text annotation (feeds all areas)
- large annotated corpora
 
21 Feb 2007  21:45Generated by HTML_ToPDF at rustyparts.com - 4 -
Notes
