|
01-14-2004 Notes-a
|
IAMTC
|
|
Notes for Content Committee (mini-)teleconference
|
Meetings
|
Back to weekly teleconference page
Content Committee Meeting
Date: Wed January 14, 2004
Present: Bonnie, Lori, Nizar, Owen, Steve
Discussed issues:
- Plan for NSF Letter of Intent and ITIC proposal:
- One letter of intent to NSF submitted by David TODAY.
- In addition, forward this same letter of intent to Mary Harper, Boyan, Kathi with all PI's from IAMTC mentioned in cover letter, but say that we'll forward additional material after our meeting next week, as we are planning to submit a proposal to the ITIC initiative. (We should mention that 3 PI's names---Rambow, Helmreich, Levin---are not on the NSF-ITR due to recent restrictions disallowing two proposals for the same PI; however these PI's will be on the ITIC proposal.)
- For material that we forward next week, we should just send our paper from the NAACL workshop and/or material from original 2-page blurb that David wrote.
- Named entities
- Problem
- Need a procedure to handle Proper Nouns such as “Air France”, “British Airways” and “The Bank of America”
- Considerations
- Automatic recognition of named entity clusters can be done easily
- Work on named entity might not be novel… although there is always potential for contribution.
- What should the IL0 look like? And what should IL1 look like?
- Evaluation can get complicated if annotators are allowed to change tree structure.
- Options Discussed
- Easy proposal: postpone named entity handling like co-reference and other phenomena
- IL0 format should be (a) right branching with all words having POS = PN or (b) the named entity nodes are merged into a single node. (a) and (b) are really notational equivalents. Who will create IL0 from parse?
- IL1 will (a) assign omega concepts to every component of a named entity or (b) assign a special concept “NAMED-ENTITY” to each component or (c) assign said concept to the merged named entity node. In cases (b) and (c), this can be done automatically.
- Recommendations
- The committee recommends:
- IL0 named entity nodes are merged into a single node. This is done in the parser (pre Tiamat).
- IL1 named entity merged node is assigned a special concept “NAMED-ENTITY”. This can be done automatically too.
- Strongly governed prepositions
- Issue
- There are three types of prepositions dominated by a verb:
- Closely related prepositions, as in “X descrimated_against Y”
- Particles in phrasal (particle) verbs, as in “X cut Y off”
- Modifiers
- Considerations
- Form in IL0 and form in IL1
- Omega specifies phrasal verbs and some closely related prepositions with verbs… but not everything.
- take_care is there, but not look_after or look_foor
- Resolution
Version 7, Wed 14 Jan 2004 18:44:56 [NYH] - created Wed 14 Jan 2004 13:07:28 [BJD]