Annotators' Overview of the ITR Grant

1
Annotators' Overview of the ITR Grant

• Information Extraction from the Biomedical
  Literature

2
Goal

• qualitatively better methods for automatically
  extracting information from the biomedical
  literature

3
Example of eventual goal

• Program input
• Amiodarone weakly inhibited CYP2C9, CYP2D6, and
  CYP3A4-mediated activities with Ki values of
  45.1--271.6 µM.
• Output database entries meaning
• amiodarone inhibits CYP2C9 with Ki45.1--271.6
• amiodarone inhibits CYP2D6 with Ki45.1--271.6
• amiodarone inhibits CYP3A4 with Ki45.1--271.6

4
IE techniques

• high-accuracy parsing (Initial annotation
  work)
• shallow semantic analysis (Later annotation
  work)
• integration of large volumes of diverse data

5
Initial focus topics

• drug developmentwith the Knowledge Integration
  and Discovery Systems group at GlaxoSmithKline
  GSK
• pediatric oncology with the eGenome group at
  Children's Hospital of Pennsylvania CHOP

6
Develop and test

• new general methods for information extraction
  IE from text, based on ongoing Penn research in
  corpus-based
• parsing (Treebanking)
• predicate-argument analysis (Propbanking)
• reference resolution (coreference annotation,
  named entity tagging)

7
Human and machine annotation Year 1

• defining, organizing and implementing various
  kind of (human) annotation of biomedical texts
• creating (mostly by statistical methods)
  automatic analyzers to apply these same kinds of
  annotation, especially to improve productivity of
  human annotators

8
Deliverables (1)

• All in biomedical domain
• Large corpora
• Part of speech (lexical)
• Treebank (syntactic)
• Named entities (terminological)
• Propbank (shallow semantic)
• Factbank (entities and relations)
• Lexicons and tools

9
Deliverables (2)

• abstracts and full-text articles annotated with
  entities and relations of interest to researchers
  (Factbanks), such as
• enzyme inhibition by various compounds
• genotype/phenotype connections

10
Deliverables (3)

• broad-coverage lexicons and tools for analysis of
  text

11
Summary Types of annotation

• Part of speech (POS)
• (required for automated Treebanking)
• Syntax (Treebank)
• Terminology (Named entities)
• Shallow semantics (Propbank)
• Entities and relations (Factbank)
• now in progress starting June 2003

12
Expectations Time

• Work week
• Max. half-time 20 hours
• Min. 10 hours
• Weekly meeting
• These may be modified for particular
  circumstances.

13
Expectations Pay

• Starting 10/hr and up, depending on background
  and task
• Increasing with experience and productivityThis
  includes getting out of introductory training and
  into production.

14
Expectations Other

• Training and studyLearning how to do it is work
  and takes effort, not just looking at the manual
  and coming to meetings.
• Discussion
• With each other
• With leaders