A New Approach for Knowledge Management and ... - Wseas

More documents

Recommendations

Info

WSEAS TRANSACTIONS onINFORMATION SCIENCE and APPLICATIONSGiulio Concas, Filippo Eros Pani, Maria Ilaria Lunesuand its objects are its texts, essential metadata(author, title, language, publishing date, etc.) mustbe deducted and formalized from their semanticcharacteristics. Some of those metadata may befurther specified according to a hierarchicalstructure: for example, the metadata “author” maybefurther refined as “main author”, “illustrator” and,"curator".4.2 Formalization of Linguistic AnnotationsThe information in the corpus are organized asinformal annotations and the most efficient way touse their information is formalizing this annotationthrough metadata schemas. In this way, not onlyannotations can be associated to their texts, but theycan also be used as search parameters for findingtexts.Linguistic annotations created with specialsoftware, like PRAAT [20] for audio files, aregenerally stored in a semi-structured manner. Infact, each annotation is distinctly represented insidethe file, according to a defined, repetitive structurewhere the annotation texts is paired with the instantor the time interval it refers to. Moreover, thebelonging of each annotation to a certain linguisticlevel is clearly stated in the file. The formalizationof annotations in a metadata schema can beachieved using a bottom-up or inductive reasoning.Starting with the analysis of the structure of eachannotations in the file and applying inductive logic,a "category" is abstracted from every linguisticlevel. This formalization allows for easily codingand representing of annotations though markuplanguages like XML, because their structure can bedescribed with tags or markers, for metadata andtheir qualifiers, inside which a linguistic label isfound. All annotations in the same linguistic level,e.g. phonetics, can be formalized in the XML asdifferent occurrences of the same metadata called"phoneme", whose value can be made up of twoterms: linguistic label and eventually time interval.4.3 Choosing a Metadata SchemaThe use of both a deductive and an inductiveapproach allows metainformation and linguisticannotations to be formalized in a single structuredmetadata schema. Entering metadata in a knowledgemanagement system requires the selection of anoperational criterion based on the particular needsthe system has to work with.Most archives use Qualified Dublin Core as mainschema for indexing and displaying metadata andSimple Dublin Core to show them through the OAI-PMH standard. There are four main criteria forchoosing a metadata schema, with differentapproaches in metadata organization: 1) mapping ofnative metadata on existing DC elements; 2)mapping of native metadata on DC elements andcreation of new customized qualifiers for DCelements; 3) creation of a customized metadataschema, identical to the native metadata set; 4)creation of DC metadata records as abstraction ofnative metadata records and entering of the latter asattachments to the resource. Out of the criteriamentioned above, the first one is the leastsatisfactory for preservation and reuse of descriptivemetadata of resources, while the third one is themost preserving of the integrity and granularity oforiginal metadata but needs great efforts for thecreation of a customized metadata schema, togetherwith high maintenance costs for the archive. Thesecond and fourth criteria combine preservation andgranularity needs with archive management costsbetter than the other two. Choosing between themdepends solely upon the particular requirements ofthe archive.Once the decision on which criterion to use issettled, the archive must be configured so that it iscompatible with the approach of choice for metadatamanagement. In particular, if the second criterion isadopted, the DC schema must be updated with new,customized qualifiers; if the third criterion ischosen, the entire metadata schema created ad hocmust be entered into the system. In this way,customized metadata and qualifiers can be used todescribe texts of the corpus inside the archive.Generally, metadata schemas can be configuredthrough the user interface of the archive. However,schemas rich in elements and qualifiers are betterconfigured with the import tools provided bymanagement systems, after having encoded themwith the XML markup language. XML is used byarchives to manage the import-export of metadata.Compilation of metadata records associated totexts in the corpus may be usually done with either auser interface or with batch import tools. Instead,when big quantities of metadata need to beassociated to one resource, like with linguisticannotations, there are specific batch import toolsthat require the specification of all metadata asattribute-value pairs, coded in an XML file.E-ISSN: 2224-3402 142 Issue 5, Volume 10, May 2013
WSEAS TRANSACTIONS onINFORMATION SCIENCE and APPLICATIONSGiulio Concas, Filippo Eros Pani, Maria Ilaria Lunesu5 Case studyThe ASAS project (http://asas.flosslab.it) aims tocreate an IR with an annotated spoken languageelectronic corpus that could become a platform forthe preservation, study, communication andappreciation of oral traditions of the Sardinianlanguage, especially improvised poetry.5.1 Annotations through PRAATThe electronic corpus was annotated by linguistsand musicologists through the PRAAT software[19], which, besides performing spoken languageanalysis, allows for multilevel segmentation andlinguistic annotations of audio files. The softwarehas a graphic interface with waveforms and voicespectrum that make annotators' work easier andmake visible those acoustic phenomena that can befound by an accurate spectrum analysis, followed byannotation levels. Linguists and musicologistsworking on the Sardinian Linguistic Sound Archivechose a list of possible annotation levels (syllable,tone, morpheme, syntagm, accents, etc.), useful forboth linguistic and musical analysis of audiorecordings.5.2 Metainformation Associated to AudioRecordingsMusicologists and Linguists, other than withannotations, wanted to complete every audiorecording by describing it with a number ofinformation, chosen among the most relevantfeatures of the recordings. The information could beused to manage recordings in the archive, becauseby describing them they allow for selection andorganization, facilitating efficient retrieval andusage. Metainformation range from somethingclosely related to cataloguing, like author, title,object, recording date, etc., up to more technicalinformation like the different singing types, speechtypes, accompaniment or instruments. Linguists andmusicologists selected 38 metainformationsassociated to audio recordings: title, author, object,description, format, etc.characteristics of texts. Through a continuousdialogue with the scholars, audio recordings wereanalysed for their essential and basic properties,needed to organize and retrieve texts in the corpus.Twelve general metadata were found: title, author,publisher, object, contributor, date, place, occasion,document accessibility, language, description andformat. Those metadata outlined the necessaryinformation to describe spoken texts in the corpus,conveying in particular singing or speech type, theoccasion in which the audio was recorded, and thelinguistic variety it belongs to.The top-down approach proceeds to furtherspecialize the metadata. More specific, or qualified,metadata are represented by adding a qualifier to thename of the more general metadata and using thecommon syntax metadata.qualifier.Lastly, "relational" metadata are defined as well,in order to define a certain relation among two ormore different objects belonging to the corpus. Aninclusion relation must be specified in order todescribe the belonging of one or more objects to thesame recording set, for example different songs in asinging contest.5.4 Formalization of Linguistic Annotations:Bottom-Up ApproachThe formalization of annotations in a metadataschema can be achieved using a bottom-up orinductive reasoning, as explained in the previoussection. The structure of annotations is analysedwith the PRAAT software. Annotations areorganized with a precise structure: each annotationis made of a time interval and a text label or by aninstant and a marker with its text. All annotations inthe same linguistic category are collected in thesame tier (or annotation level), which can beconsidered as the category they belong to, giving itsname to the corresponding metadata. In this way, arepeatable metadata is found in each annotationlevel of the TextGrid (the text file where PRAATstores all Tier with their own segmentations andannotations) and each annotation can be representedas multiple occurrences of that metadata.5.3 Formalization of SemanticCharacteristics: Top-Down ApproachAfter designing the conceptual model of theknowledge domain, a top-down or deductiveapproach can be used for formalizing the semantic5.5 Choosing a Metadata Schema for KMSEnteringDepending on the interoperability needs that mustbe met, importing the metadata schema that was justcreated into the knowledge management systemE-ISSN: 2224-3402 143 Issue 5, Volume 10, May 2013
Page 1 and 2: WSEAS TRANSACTIONS onINFORMATION SC
Page 3: WSEAS TRANSACTIONS onINFORMATION SC

A New Approach for Knowledge Management and ... - Wseas

Create successful ePaper yourself

Delete template?

Save as template?