Musical Works and Information Retrieval

Abstract

Music bibliographers and librarians are accustomed to an essential dichotomy that presents itself in the ordering of musical documents for retrieval. The documents themselves (scores and recordings, mostly) must be represented in retrieval systems according to their physical characteristics (that is, according to formatted physical descriptions) for the purposes of general bibliographic control; but the musical works that are contained in and conveyed by these documents require representation according to their intellectual origins and musical characteristics (that is, author and uniform title headings). Collections of musical documents are unique among collections of documents (in libraries, bibliographies, etc.) in that the influence of repertory is such that a given collection will have many instantiations of the same musical work--one Tchaikovsky Fifth Symphony (a musical work), but a dozen scores of different sizes and formats, and dozens of recordings, not to mention excerpts and arrangements. Thus the intellectual control of musical works assumes tremendous importance in the work of music librarians and bibliographers. At the beginning of this new century we find a rapidly evolving domain for research and very pragmatic implementation known as Music Information Retrieval (MIR). Music Information Retrieval is basically the activity of automating the retrieval of musical works, or parts of musical works. MIR embraces everything from query-by-humming systems that allow a searcher to hum a tune for which the database returns an audio output, to the design of metadata structures and standard name-title-subject querying of bibliographic databases. The field encompasses various system-Engineering components, such as audioinformation retrieval (AIR), and it also encompasses musicological engineering problems such as recognition of parts of musical works. A major component is the creation of digital music libraries--electronic repositories of musical works. We will look more closely at the components of MIR later, but for now it is important to comprehend the evolution of this new domain for music librarianship and bibliography. In information retrieval in general, as well as in music information retrieval, the objects or that will be mapped into databases for retrieval must be well-defined. An entity is simply a thing--it might be a physical characteristic or it might be a concept. Entities occupy prominent positions in the structure of information retrieval databases. Consider an online catalog: therein things such as tides, composers' names, tide-page transcriptions, etc., could all be considered entities. Entities have attributes (inherent characteristics that can be defined) that are used to structure their representations in information retrieval systems. In music information retrieval, as we will see below, there are many potential entities at play. But universal to the retrieval of music information--whether it be bibliographic or audiographic using surrogates or real-time sound--is the concept of the musical work. Musical works, then, as opposed to musical documents such as scores or recordings of musical works, form a key entity for music information retrieval. Ultimately, searches for a given musical work rely on the hope of subsequent selection of instantiation in one of several documentary formats. Musical works have been variously and industriously described by musicologists and music bibliographers. Yet, in the information retrieval domain, the work as opposed to the document has only recently received focused attention. (1) Efforts to define works as information retrieval entities and to document their occurrence empirically are quite recent. In fact, systems for bibliographic information retrieval, and more recently for information storage and retrieval, have been designed with the document as the key entity, and works have been dismissed as too abstract or difficult to define empirically to take a role in information retrieval. …