ISMIR2012 late breaking session summary

A small but enthusiastic group of people attended our late breaking session at ISMIR 2012 we held jointly with Sebastian Ewert to discuss the SOVARR and Semantic Media projects. The participants included computational musicologist Tim Crawford from Goldsmith College, University of London, as well as Linked Data and Semantic Web experts like Sean Bechhofer from the Information Management Group at the University of Manchester School of Computer Science or Kevin R. Page from the Oxford e-Science research center.

We started by discussing the feasibility of creating shared ontologies within research communities such as the Music Information Retrieval (MIR) community. It was commonly agreed that the creation of a single shared vocabulary is a difficult challenge and a more feasible alternative is exploring the possibility of several shared modules that are more domain or task specific, that is, vocabularies that serve a specific task in music information research such as genre classification, or more fine tuned to a specific tool or a part of the community with specific research practices. These modules could potentially be linked on a higher level or be rooted in a dynamic, community authored vocabulary. Semantic Web ontologies have properties that are ideal to support our requirements. It is possible to create flexible and modular systems that still support the unique identification of terms, the possibility of establishing hierarchical or equivalence relationships between them, and describing the meaning of data at different levels of detail.

As a possible development path, Kevin Page suggested the examination of particular MIR tasks or workflows, such as the ones run by the Music Information Retrieval Evaluation eXchange (MIREX). This could lead to creating vocabularies and ontological models that support particular workflows. Demonstrators that use shared open vocabularies in this context can greatly facilitate their adaptation.

An example of such a demonstrator was previously built during the Networked Environment for Music Analysis (NEMA) project. The Country/Country demo asks the question “How country is my country?”. It demonstrates the utility of ontologies, such as the Music Ontology, and Linked Open Data in a typical MIR workflow involving data collection and publishing as well as signal processing in audio based genre analysis. A video of this application where Tim Crawford, Senior Lecturer in computational musicology at Goldsmiths University of London, walks through a common use case in the proof-of-concept system is available here. Sonic Annotator Web Application (SAWA) is another demonstrator that was built to help researchers to learn about a previously created ontology for audio features.

Rudolf Mayer from the Institute of Software Technology and Interactive Systems (ISIS) at the Vienna University of Technology suggested to look at advancements in the Digital Preservation community. This field was originally concerned with the preservation of static digital objects such as multimedia documents to make them resilient to the rapid changes in storage and information access technologies. An emerging topic in digital preservation is the preservation of complex dynamic digital objects such as workflows in scientific research. A couple of presentations about this field is available here. In their recent ISMIR paper Mayer and Rauber argue that previously proposed benchmarking environments, such as MIREX and decentralised frameworks developed in other research communities, do not support the documentation of process execution during experiments. The idea of sharing reusable workflows is an important motivating use case for creating shared vocabularies. Therefore use cases arising in the preservation of these workflows are indeed important to consider in the SOVARR project.

An interesting idea and possibility emerging from the above requirements is the creation of a web-based service where members of the research community could register their audio feature extraction techniques. The registered methods would receive a unique identifier, and they could be linked with computational algorithms, publications and other relevant metadata. Such a service could also be the basis for resolving terminological differences, since equivalence relations could easily be established to state that features named differently in different tools but follow the same computation steps are indeed the same. However, it remains difficult to decide who should have authority over creating such links.