110 likes | 201 Views
Depositors’ usage of IMDI metadata. Daan Broeder & Alex Klassmann MPI Institute for Psycholinguistics. DELAMAN meeting London 2006. IMDI metadata. Forms with ~150 possible descriptors Describes bundles of related resources Extensive set compared with DC/OLAC
E N D
Depositors’ usage of IMDI metadata Daan Broeder & Alex Klassmann MPI Institute for Psycholinguistics DELAMAN meeting London 2006
IMDI metadata • Forms with ~150 possible descriptors • Describes bundles of related resources • Extensive set compared with DC/OLAC • But only “name” descriptor is compulsory • Archive holds • ~40000 IMDI sessions or resource bundles +15000 non-local but available in our DB • Describing ~150000 resources
IMDI Metadata The descriptors hierarchically ordered entries, which concern • the event (recording location, date, etc), • the project, • the languages involved, • the Participants, • the type and nature of speech, • technical information about the resources • access rights • values of descriptors can be closed or open vocabularies or free text. • user can use prose descriptions at each of these levels + project/user defined keys
Metadata Use • Documentation of the resources • Retrieval and reuse: archive offers tools for: • Browsing the archives’ corpora • Structured metadata search • High precision, low recall • Unstructured google-like metadata search • High recall, low precision • Large set-> not all elements are always relevant • Sparsely populated metadata space • Search tool to show frequency counts for metadata values. Avoids fruitless searches.
Depositor Guidance • In general depositors are urged to be complete as possible for documentation purposes • Some projects have an obligatory set of descriptors to fill in. (CGN, DBD, …) • Provide training to get familiar with the set and tools • Provide documentation • Support by student-assistants and corpus managers
Observations II • Often researchers do not fill in all the relevant data at their disposal. • Some tendency to avoid this time-consuming work oriented to re-usage by others. • The sheer size of the set may discourage people to start filling in data at all. • Training helps. • Best results in projects that decided beforehand what descriptors were needed to fill in. • Of course there are also very committed individuals!!! • Corpus managers/student assistants may clean things up. • but limited use since only the researcher has specific knowledge • can serve as intermediaries.
Observations II • Only that part of the archive where metadata was specified manually (e.g. CGN was excluded as were sessions outside the MPI) • Statistics on the basis of ~25000 remaining sessions • The data gives an impression of how often fields are actually filled in (e.g. not empty and not default “unknown“ or “unspecified“). • Cannot exclude “repairs” where obvious omissions were repaired by corpus management
Descriptor name total-25000 fl-12000 acqui-10000 • Country 93 93 99 • Address 15 21 15 • Region 7 10 11 • Description 48 30 77 • Key 33 17 58 • Project.Name 90 91 87 • Content.Description 93 95 97 • Genre 29 44 15 • SubGenre 23 34 13 • Task 43 49 34 • Modalities 80 80 82 • Subject 3 6 2 • Interactivity 73 72 81 • PlanningType 53 51 73 • Involvement 70 71 72 • SocialContext 6 10 9 • EventStructure 7 9 9 • Channel 8 10 11 • Content.Language.Description 43 25 67 • Content.Language.Id 91 90 91 • Content.Language.Name 91 90 94
Actor.Language.Description 33 14 61 • Actor.Language.Id 25 20 53 • Actor.Language.Name 47 37 83 • Actor.Role 94 97 99 • Actor.Name 94 95 99 • Actor.FullName 90 93 97 • Actor.Code 70 68 84 • Actor.FamilySocialRole 24 31 18 • Actor.EthnicGroup 14 20 13 • Actor.BirthDate 5 8 8 • Actor.Age 44 47 50 • Actor.Sex 70 69 92 • Actor.Education 13 16 11 • Actor.Description 65 78 56 • Actor.Key 52 44 68 • MediaFile.Type 85 83 85 • MediaFile.Format 85 83 85 • MediaFile.Quality 18 8 31 • WrittenResource.Type 67 57 71 • WrittenResource.SubType 30 19 35 • WrittenResource.Format 56 42 70 • WrittenResource.ContentEncoding 3 7 0 • WrittenResource.CharacterEncoding 3 12 0 • WrittenResource.LanguageId 4 1 1
Conclusions • As can be seen the sets are far from being complete. • But also every field of the scheme has been used in some sessions, so that it seems that no field in the schema is obsolete • People find use for the description fields that are available at different levels (~50%) • Also the user/project defined keys are used (~50%) -> IMDI set is not big enough • Some keys are not much used • Remove? • But where then to put this information if its available?