|
11-14-2003 Notes
|
IAMTC
|
|
Notes from 11-14-2003 Teleconference
|
Attendees: Ed H, David F, Steve H, Owen R, Nizar H, Simon, Keith M
Topics
Note: We should prepare agenda ahead of time from now on (host responsible).
- Workshop (Ed):
Adam M.'s workshop was accepted. He wants more tradidional model. We can submit proposal for additional half day which we will control, and where we will have invited speakers, proposals/questions from the audience, and suggestions for collaboration. This is being proposed to Bonnie now, and there is possibility to scale back, either to single session or to just a panel (Owen's suggestion).
This would be in addition to workshop at ACL, which could be tailored more to international funding perspectives and interests. Ed will put together and circulate a proposal for this workshop. We should identify Europeans who would also like to participate.
- status of texts
- Keith will move 2 English translations of French text to obvious location
- These will be two English transaltions of the same French text, so we can start comparing differences/divergences in addition to refining the annotation standard.
- parser / tools
- currently designing annotation tool (ISI); programming will begin today or soon thereafter. Ed requested parser from Owen. Bonnie and Nizar will send verb frame elements. Status is that someone is working on this right now (the interface between WN sense with set of roles appearing with verb, and mapping between roles and syntactic relations, along with example and definition for each). One issue is that the annotation committee has not come to firm decision on sets of roles; however, as Owen suggests, the real issue is ease of use for assuring interannotator agreement. Should be possible to map to different role names in future, if necessary. This could be based on, e.g. Levin classes. Note the similarity to PropBank
- Owen is working on getting parser ready for distribution. Ed suggested potentially running as a service. Owen will either get a student to get distribution ready or will do this himself.
- Question as to whether parser will be part of annotation tool or separate: Ed answered that in the ideal design, they will be integrated.
- Question as to comparison of Connexor w/ Dekang Lin's parser: we could look into this, but this would be a lot of work to do now. There are some licensing restrictions on DL's code. Minipar is free, without the clustering, however (according to anecdotal evidence, anyway). UMD student has some experience with this for LCS composition; Nizar will follow up with Owen.
- Output format for parser will be one line per node, with converter for input to GRAPH.
- What features do we want annotators to be able to see in popup windows? Should be able to look at least at what they have done/added. Do we want syntactic features like voice, rel, etc...? We need to know this, because they need to be in the header of the file input to GRAPH, since this specifies which features are displayed/available.
- Platforms: Linux/Solaris ... ? other?? I missed this part due to updating the Wiki... can someone fill in here, please?
- First interface to Omega will be like public web interface
- For now, type in word, click "Word", get number of possible entry points to ontology, can traverse ontology
- annotation manual
- tools will likely be ready before the manual
- for verbs, it looks like mostly just taken directly from Owen's dependency manual
- how is test annotation cycle tied into annotation manual development and refinement
- copulas and prepositions: discussions not restarted yet - Owen will initiate
- phrasal verbs: we can use PropBank - of PropBank combines V+P, they are combined, if not, then separate
- start with manual as it is posted on wiki page
- can start with Bonnie and Nizar's first cut at roles. But roles must not be chosen freely - they are chosen in conjunction with verb senses.
- annotators
- NMSU: 2 annotators on board, will start formal process for others (will take 1 month, but for more permanent)
- MITRE: several potential identified, near-native French, and native English
- Colombia: will discuss hiring process today. Question: Is is still alright to have more than one person in the role of a single annotator? This is not ideal due to main project goal of high interannotator agreement. Colombia will endeavor to
- CMU: Teruko is looking at some Japanese NS annotators, Lori, Teruko, and Simon will be annotators for English. There is some discussion of this. We could continue discussion of division between researchers and annotators here if desired.
- UMD: Hired Arabic NS, fluent in English. Having trouble finding English NS. Keith may be able to help if necessary.
- ISI: hard time finding English. May use Korean NS, who are very fluent in English. Also, looking at hiring someone from NMSU as "Ontology Czar".
Rough Schedule
We also discussed a Rough Near-Term Project Schedule. Please take a look and update as necessary.
Version 1, Wed 26 Nov 2003 15:05:42 [KJM]