Changes between Version 4 and Version 5 of Taskforces/FCS/VidConf20150624
- Timestamp:
- 06/24/15 13:25:38 (9 years ago)
Legend:
- Unmodified
- Added
- Removed
- Modified
-
Taskforces/FCS/VidConf20150624
v4 v5 14 14 15 15 == Outcome: Minutes, Decisions and Actions (currently draft) == 16 Main focus Dataview return format16 Main focus of today's Dataview return format 17 17 18 Not much examples except for Oliver's. 18 Not much examples except for Oliver's. We need more. 19 19 20 Advanced dataview 21 Adhoc or based on something? 22 Oliver mental model one dataview will convey all layers etc 20 Matej, are these exampel formats for Advanced dataview adhoc or based on something? 23 21 24 So no msd dataview 25 explict elements for some layers 26 container format. 22 Oliver: mental model of that one dataview will convey all layers etc. E g no msd dataview or explict elements for some layers. It is a container format. 23 27 24 Pavel, looks nice. Reason behind it? 25 28 26 Any tree? Complex tree view? Multiple layers? 29 27 30 Press stop 2-3 secs after stop talking. 31 Hierarchical structure. 32 Proper scope of work. 28 Press stop broadcasting 2-3 secs after you stop talking. 29 30 == Questions on Hierarchical structure == 31 32 We need to define a proper scope of work. 33 33 Transport hierarchy 34 34 Matej, really another complexity level. Start with query annotation layers, and only when we master that move forward. 35 35 36 Pavel, second suggestion multi annotations. Fine with forgetting trees for now. structured/hierarchical attributes.37 I think it can be nice in terms of ease of understanding that some thing belong together.36 Pavel, second suggestion multi annotations. Fine with forgetting trees for now. Structured/hierarchical attributes. 37 I think it can be nice in terms of ease of understanding that some things belong together. 38 38 39 Matej, what do you mean? Example.39 Matej, what do you mean? Provide example. 40 40 41 41 case 42 42 43 Oliver (IDS): question of course is, if structured, how would it be structured ;)43 Oliver: question of course is, if structured, how would it be structured ;) 44 44 a client like the aggregator needs to make sense of that 45 45 46 harmonize if they are structured. 47 No good solution yet. 46 We need to harmonize if they are structured. No good solution yet. 48 47 49 48 Dieter: Pavel: is there a standard way for that in the universal dependencies? 50 49 51 Controlled vocab 52 Like for POS the UD-17 we decided on. 53 Jörg not every endpoint will provide UD-17 we are not going to retag all corpora. 50 Matej: If we go this route we need to come up with even more controlled vocabularies, like for POS the UD-17 we decided on. 54 51 55 Pavel, agree hierarchical attributes is too complex for now. 52 Jörg: not every endpoint will provide UD-17. We are not going to retag all corpora. 56 53 57 Jörg bidrectional translation. 54 Pavel: agree hierarchical attributes is too complex for now. 58 55 59 Dieter, limited list 60 choice 61 first proposal, syllables. 62 finer granularity. In worst case every single character. 56 Jörg: bidrectional translation is needed. 63 57 64 Candidate for speech 65 or primarily for textual resources? 58 Dieter: limited list layers 59 before making a choice 66 60 67 time lined signal 68 "textual" annotations 61 == Question about first proposal, syllables == 62 Oliver: Finer granularity. In worst case every single characters. 69 63 70 Dieter concrete example. Silence is best covered by example 1. 64 Dieter: Candidate also for speech or primarily for textual resources? 65 66 Basically time lined signal "textual" annotations 67 68 Dieter: provides concrete example. Silence seems to be best covered by proposal 1. 71 69 Transcribed speech corpora. 72 70 73 Matej , generic way to describe atomic units. So time stamp could be fitted with example 2.71 Matej: Not really, proposal 2 is better but needs other unit like time stamps. Generic way to describe atomic units. 74 72 75 Oliver ,yes, offsets can well be timestamps73 Oliver: yes, offsets can well be timestamps 76 74 77 silence background noice75 silence, background noice are interpretations/computed annotations 78 76 79 Combination of dataviews 80 allow to reference other layer items 81 non-superior layers 82 order is not defined, but 77 What about combination of dataviews 78 allow to reference other layer items. This was disussed at the workshop as one alternative LJO wants to provide proposal. 79 80 Oliver: non-superior layers, order is not defined in the current proposals 81 83 82 Dieter: I have the impression we are re-inventing formats like EAF: https://corpus1.mpi.nl/media-archive/demo/Ams_Demo/versioning_demo/Annotations/118_fishing2-fire-2011.eaf 84 83 85 Comparison matrix. 86 87 Keep things simple. 88 89 Keep it fairly straight forward. 90 91 Keep non-textual formats in mind. 92 93 Doing concrete proposals. More are needed. 94 95 transform to standard format. 96 97 What about parallel corpora? 98 another annotation layer trans 99 100 More discussion is needed. Concrete proposals. 101 102 Hanna, concrete examples? Propose new layers? 103 Mappings are welcome. 104 105 Use second view in examples 106 107 Need attribute for unit. 108 109 Matej, lexical resources. 110 111 Toying 84 LJO action: Add EAF to comparison matrix wehen we have more examples. 112 85 113 86 114 LJO: add page for advanced datawiew to trac. 115 Next meeting, 2-3 weeks 116 Might be flash or adobe connect 87 Keep things simple. We need to keep it fairly straight forward and still keep non-textual formats in mind. 88 89 We need to work on getting more concrete proposals available. More examples are needed. 90 91 The transport format and its data could be transformed to standard format if needed. 92 93 == Question about parallel corpora == 94 Adds another annotation layer, potentially translation also 95 96 Consensus for actions: 97 * More discussion is needed. 98 * Concrete proposals. 99 * concrete examples 100 101 Hanna: propose new layers? Mappings are welcome. 102 103 Decision: Use second proposal for new examples or another proposal with examples 104 105 Diter: Need attribute for unit. 106 107 108 == Question about lexical resources == 109 Matej raises the question about how to handle lexical resources. 110 111 112 == Actions summary == 113 * ALL: Toying with the current proposals and example 114 * ALL: Provide proposals 115 * ALL: Provide examples 116 * LJO: add page for advanced datawiew to trac where these can be put for reference, still important to use mailing list for discussion 117 * Next meeting, 2-3 weeks, might be flash again or adobe connect 117 118 118 119