Toward a format-neutral annotation store. (September 2017)

• Record Type: Journal Article
• Authors: Fromont, Robert
• Is Part Of: Computer speech & language. Volume 45(2017)
• Page Start: 348
• Page End: 374
• Publication Date: 2017-09
• Subjects: Speech corpora -- Languane annotation -- Annotation graph -- Interoperability; Speech processing systems -- Periodicals; Automatic speech recognition -- Periodicals; Computers -- Periodicals; Linguistics -- Periodicals; Speech-Language Pathology -- Periodicals; Traitement automatique de la parole -- Périodiques; Reconnaissance automatique de la parole -- Périodiques; Automatic speech recognition; Speech processing systems; Electronic journals; Periodicals; 006.454
• DOI: 10.1016/j.csl.2017.01.004
• Languages: English
• ISSNs: 0885-2308

Abstract

Highlights: Speech corpora are difficult to share and reuse because of format incompatibilities. Various central 'pivot' models for annotations have been proposed as a solution. LaBB-CAT uses Annotation Graphs with extensions to formalise annotation structure. An API for the resulting model can be used for query, annotation and conversion. Abstract: Sharing speech corpora and their annotations is desirable, in order to maximise the value gained from the expense and hard work involved in transcribing and annotating them. However, differences in conventions and format are barriers to sharing of data; text conventions conflict, file formats differ, and annotation ontologies do not match up. Using a 'pivot' form to store annotations in a tool and format neutral manner can alleviate many of these difficulties. There are several possibilities for the pivot form, including the Annotation Graph model, which meets most of the requirements to be a pivot. The LaBB-CAT software's implementation of Annotation Graphs incorporates some extensions to the model, which handle the remaining unmet requirements, and create the possibility of defining an annotation API that makes automation of conversion, querying, and manipulation of annotations easier.