BerlinDissemination

Availability of DELPH-IN Tools for Online Demos, Dissemination, and Modes of Cooperation among the DELPH-IN Partners

Moderators: LarsHellan and HansUszkoreit

Scribe: UlrichSchaefer

Thu 23, Aug 1135-1300

Tools for Online Demos

LH:

• for presentation to outside world and among grammar writers
• testsuite, document covered phenomena
• selected demo sentences showing coverage of grammar
• online interface
• display "small feature" structure with result
• morphology, syntax, semantics
• show details details interesting for MT/IE

OE: market to community, have more semantics, show less? BC: parse selection not shown in LKB interface US/BC: ERG+GG online demo only displays tree + MRS AC/DF/OE: treebanked parse selection possible LH: online, BC: more for internal, development purposes (co-developers) VK: and for teaching

DF: show simplified, flattened result, tools exist, partly grammar-specific AC: XSLT not for linguists, CSist necessary in addition OE, LH: avm button , configurable filtering important, but end-user or grammar-writer specific. AC: display sub-FS under paths AC/LH/OE: not only hiding sub-FSes, but also displaying differently OE: xslt / stylesheets to display in browser OE: what tools exist AC: TFS DTD, ISO, could be implemented in LKB US: applet in heart of gold, DTD OE: mori-5 tsuji HTML rendering -> ADD LINK YM: less for linguists US: see table in these minutes for online examples

GG, ERG, www, tsuji, sproutfs (DTD)

Modes of cooperation

• HU: applications
  • groups that are interested in / working on Information Management, IE, event, relation extraction: tokyo, cambridge, saarbrücken, oslo, melbourne
• AC: paraphrase task (as proposed before): interesting task with nice properties:
  • take existing machinery to showcase (using parser + generator + some simple idioms)
  • types of phenomena in paraphrasing (example): light verbs
  • demo system: type in sentences, let system paraphrase, let people judge whether OK or not to get data
  • set as common task?
  • IE in chemistry/medical domain too special as common task
• HU currently event/relation detection corpus-based
• if with deep processing: make sure that it works with paraphrasing
• paraphrasing as support for MT
• AC: learning paraphrase is learning semantic equivalences
• also in grammar, style checking
• JC: text to text paraphrasing, summarization, compression as generic applications, too generic?
• AC: rewrite rules; learn or hand-write
• HU: entailment
• HU, AC: delph-in as forum for seeing which groups are interested
• learn about preferences of the groups (e.g. dialogue)
• interest in dialogue, question answering:
  • Lisbon
  • Saarbrücken
  • Trondheim

OE: research interest (also) depend on funding opportunities delph-in goal: get funding (depends on whether application-oriented or scientific vision)

HU: delph-in as a whole did not and (because of missing formal frame) can not get funding (compared to ACE, CLEF), but groups individually got funding with the help of delph-in in background, e.g. lisbon, saarbrücken (PIRE)

gain reputation and credibility

bilateral grants (NSF, BMBF) possible within delph-in partners

AC: what could delph-in do specifically to help people get funding?

Dissemination

HU:

• common evaluation tasks
• connection with (larger) companies, SME
• visibility at international conferences (thank delph-in in papers)

AC:

• delph-in web page, mention on homepage of members
• automatically grasp

official maintainer of delph-in web page

OE: facelift, wikify

AB: Lisbon got a second funding, almost maximum in Portugal, thanks to delph-in support

Google visibility: pargram vs. delph-in

web page: color scheme, content, news, publications

cf. Fefor