12th International Society for Music Information Retrieval Conference (ISMIR 2011) KNOWLEDGE REPRESENTATION ISSUES IN MUSICAL INSTRUMENT ONTOLOGY DESIGN Sefki Kolozali, Mathieu Barthet, Gyorgy¨ Fazekas, Mark Sandler Centre for Digital Music, Queen Mary University of London, London, UK {sefki.kolozali, gyorgy.fazekas, mathieu.barthet, mark.sandler}@eecs.qmul.ac.uk ABSTRACT musical instruments, for instance, (ethno)musicologists have been working on creating a common vocabulary, This paper presents preliminary work on musical in- which represents all instruments with relevant charac- struments ontology design, and investigates heterogene- teristics in a systematic way. The classification of in- ity and limitations in existing instrument classification struments has also been investigated by organologists schemes. Numerous research to date aims at represent- and museologists [8]. Hornobostel and Sachs [14] pro- ing information about musical instruments. The works posed a musical instrument classification scheme as an we examined are based on the well known Hornbostel extension of Mahillon’s scheme [9], originally designed and Sach’s classification scheme. We developed repre- to catalogue the worldwide collection of musical instru- sentations using the Ontology Web Language (OWL), ments housed in the Brussels Conservatory Instrumen- and compared terminological and conceptual heterogene- tal museum. ity using SPARQL queries. We found evidence to sup- port that traditional designs based on taxonomy trees The Hornobostel and Sachs classification scheme (H- lead to ill-defined knowledge representation, especially S system) relies on a downward taxonomy by logical di- in the context of an ontology for the Semantic Web. vision. The method later coined Systematik by Drager´ In order to overcome this issue, it is desirable to have [4]. Although many attempts have since been made by an instrument ontology that exhibits a semantically rich scholars to improve the Hornobostel and Sachs’ Sys- structure. tematik, it is still predominant in museums around the world. Kartomi [8] attributes the success of the classi- 1. INTRODUCTION fication system to the fact that it is essentially numer- ical rather than lexical, making it an international sys- Ontologies are used to represent knowledge in a formal tem (e.g. 211.11-922 refers to the timpani or kettledrum way. For instance, they can be used to enable machines in the H-S system). Elschek [5], was the first to pro- to make sense of the unstructured nature of informa- pose an upward method of classification based on in- tion available on the Web. Compared to simple meta- strument attributes complementing downward classifi- data encoding, ontologies provide meaning by defining cations schemes such as the Systematik. concepts and relationships in an application domain, as well as constraints on their use. Furthermore, they per- The purpose of our paper is to investigate knowledge mit interoperability, automatic reasoning and access to representation issues of musical instruments on the Se- information using complex queries. mantic Web, by taking various musical instrument clas- Knowledge representation in the domain of musical sification schemes into account. The rest of the pa- instruments is a complex issue, involving a wide range per is organised as follows: In section 2, we give an of instrument characteristics, for instance, physical as- overview of the Semantic Web standards used in this pects of instruments such as different types of sound study. In section 3, we describe the Music Ontology initiation, resonators, as well as the player-instrument and the related instrument ontologies. In section 4, we relationship. Since the 19th century, numerous studies detail knowledge representation issues of various mu- developed systems for representing information about sical instrument classification schemes, and highlight Permission to make digital or hard copies of all or part of this work for their conceptual heterogeneities. In section 5, the OWL personal or classroom use is granted without fee provided that copies representations of these classification schemes are ex- are not made or distributed for profit or commercial advantage and amined using SPARQL queries. Finally, in the section that copies bear this notice and the full citation on the first page. 6, we note on further difficulties of the research prob- c 2011 International Society for Music Information Retrieval. lem, and outline our future work. 465 Poster Session 3 2. SEMANTIC WEB TECHNOLOGIES 3. RELATED WORK Our primary aim is to develop a semantically rich on- The Semantic Web is an initiative of the World Wide tology of instruments which can be used in conjunction Web Consortium (W3C) which proposes standards un- with the Music Ontology 5 . In this section, we outline derlying the technologies of the Web [10]. The W3C in- this ontology and previously published Semantic Web vestigates how to maintain interoperability and univer- ontologies of musical instruments. sality of the Web using open standards and languages. The Music Ontology [13] provides a unified frame- The technologies relevant in our examination of issues work for describing music-related information (i.e. ed- in musical instrument ontology design are presented in itorial data including artists, albums and tracks) on the this section. Web. It is built on several ontologies such as the Time- RDF: The Resource Description Framework (RDF) 1 line Ontology 6 , the Event Ontology 7 , the Functional is a simple data model, that associates subjects and ob- Requirements for Bibliographic Records (FRBR) On- jects using a predicate. A series of connections can tology 8 , and the Friend Of A Friend (FOAF) Ontol- be made using triples or three-tuple associations, which ogy 9 . It subsumes specific terms from these ontologies, form a graph of semantic relationships. RDF is the basis useful to describe music related data. The Timeline and for more complex knowledge representation languages Event ontologies, can be used to localise events in space such as the RDF Schema Language (RDFS). See for in- and time. The FRBR model links books and other in- stance [2] for more details. tellectual works with their creators, publishers or sub- jects, and provides a model to describe the life cycle of SKOS: The Simple Knowledge Organization Systems these works. This is reused by the Music Ontology to (SKOS) 2 is a semi-formal model for expressing con- describe the music production workflow from composi- trolled vocabularies (classification schemes, thesauri, tax- tion to delivery. Finally, FOAF defines people, groups onomies) in RDF. It defines skos:Concept, whose and organisations. The Music Ontology does not cover individuals may be associated with one or more lex- every music related concept, rather, it provides exten- ical labels, skos:prefLabel, skos:altLabel sion points where a domain specific ontology, such as and placed within a hierarchy using skos:broader, a musical instrument or a genre ontology may be inte- skos:narrower, or skos:related properties, ex- grated. hibiting a thesaurus model [1]. Based on the Musicbrainz 10 instrument tree, Her- man 11 published a musical instrument taxonomy ex- OWL: The Ontology Web Language (OWL) 3 is a pressed in SKOS. This serves as an extension to the Mu- a W3C recommendation for defining and instantiating sic Ontology. While SKOS is well suited for hierarchi- web ontologies. Like RDFS, OWL permits the defi- cal classification schemes, it provides limited support nition of classes, properties and their instances, and is for other types of relationships; skos:related for used to explicitly represent the meaning and relation- example, may be used to describe associative relations, ships of terms in vocabularies, and express constraints but only in a semi-formal way, without a more explicit on their use. Such a representation is called ontology. definition. Moreover, the transitivity of broader and nar- OWL has a richer vocabulary than RDFS and SKOS, for rower relations are not guaranteed in SKOS, therefore it example, for specifying cardinality, equality, character- is difficult to infer for instance the instrument family of istics of properties such as transitivity or symmetry and a given instrument, without additional knowledge not enumerated classes. [1]. expressed in the model. While this taxonomy is suit- SPARQL: Simple Protocol and RDF Query Language able for applications that require only a semantic label (SPARQL) 4 defines a standard access protocol for RDF to represent instruments associated with audio items, it that provides Semantic Web developers with a power- is insufficient if the heterogeneity of instrument rela- ful tool to extract information from large data sets. A tions has to be explicitly represented. 12 query consists of several graph patterns, which can be The Kanzaki Music Ontology also contains a small combined recursively to form arbitrarily complex query instrument taxonomy. However, there are only 5 instru- patterns. It may be used for any data source that can be 5 http://musicontology.com/ mapped to RDF. 6 http://purl.org/NET/c4dm/timeline.owl/ 7 http://purl.org/NET/c4dm/event.owl/ 8 http://vocab.org/frbr/core/ 1 http://www.w3.org/TR/rdf-primer 9 http://xmlns.com/foaf/spec/ 2 http://www.w3.org/TR/skos-reference 10 http://musicbrainz.org/ 3 http://www.w3.org/TR/owl-primer 11 http://purl.org/ontology/mo/mit# 4 http://www.w3.org/TR/rdf-sparql-protocol 12 http://www.kanzaki.com/ns/music 466 12th International Society for Music Information Retrieval Conference (ISMIR 2011) ment families defined (e.g. string instruments, wood- wind instruments, brass instruments, percussion, and keyboard instruments), with 26 corresponding instru- ment classes. Although these works provide instrument taxonomies that can be used on the Semantic Web, there remains a need for a semantically rich ontology, which represents the heterogeneity as well as different compo- nents and aspects of musical instruments on the Web. Finally, a recently published XML-based taxonomy serves as an extension to Music XML 13 . This system departs form Hornobostel and Sachs, and proposes a classification scheme based on materials and performance mechanism, instead of the sound production mechanism.