Informetrics and Music Information Retrieval: An Informetric Examination of a Folksong Database

This study examined the informetric properties of intervallic n-grams as derived from a database of 9354 folksongs. The informetric properties of the melodies were compared and contrasted to those of traditional text. Understanding these similarities/dissimilarities can play a vital role in the creation of a successful Music Information Retrieval (MIR) system. Various styles of n-gram creation were examined some of which exhibited great promises. Finally, this paper shows how the informetric analyses can be used to give theoretical justifications for the application of traditional IR methodologies in constructing MIR systems. Background There seems to be as many approaches to developing Music Information Retrieval (MIR) systems as there are developers. Some have designed complex suites of computer tools to analyse all the varied facets of music (e.g., Huron 1991). Others have tried to automate the thematic catalogue by including incipits (i.e., beginning phrases) as part of a bibliographic record (e.g., RISM 1997). Still others have explored the idea of using sophisticated approximate string-matching techniques (e.g., Ghias et al. 1995; McNab et al. 1996). One thing that unites all of these approaches is that they have some kind of shortcoming. The more powerful analytic systems can be very difficult to use, incipit indexes leave out large amounts of music that might be of interest, and approximate stringmatching can be computationally expensive without necessarily giving better results (Downie 1997). Taking our cue from those thematic catalogues that have reduced the amount of musical information represented (e.g., Keller and Rabson 1980) we have been developing a prototype MIR system (Tague et al. 1993; Downie 1994) based upon the intervals (i.e., the signed differences between pitches) found within monophonic melodies (Downie 1995, 1997a). We believe that there is enough information contained within an interval-only representation of monophonic melodies that effective retrieval of musical information can be achieved. We aim to extend the thematic catalogue model by affording access to melodic segments found anywhere within a melody. To achieve this extension we fragment the melodies into length-n subsections called n-grams. The length of these n-grams and the degree to which we precisely represent the