‘Corpus Research as a Means of Unlocking Musical Grammar’, sponsored by the Israel Science Foundation and organized by Yoel Greenberg (Bar-Ilan University) and Uri Rom (Tel Aviv University), brought together researchers from around the world to share their ideas about corpus studies in music. This four-day workshop, co-hosted by the Tel Aviv University Buchmann-Mehta School of Music and the Bar-Ilan University Department of Music, gave scholars the opportunity to present recent research and methodologies, and to conduct an open discussion on the potential and pitfalls of corpus studies and computational methods, as well as their place in future music scholarship. The ten sessions focused on various levels of musical grammar, from the most basic structural units (pitches, rhythmic gestures), through larger syntactic elements such as harmonic progressions and cadences, to highly complex features of form and style. One assumption shared by all participants was that corpus-based approaches, now powered by computational tools and state-of-the-art statistics, can sustain a wide range of data-driven enquiries, making it possible to revisit, refine and challenge prevailing theoretical conceptions. Although the workshop covered a broad variety of topics, from plainchant to popular music and from general methodological talks to presentations of specific corpora, a particularly large proportion of the talks were devoted to eighteenth-century music. Although this emphasis may have reflected to some degree the fields of expertise of the conference organizers, it is probably also due to the suitability of this repertory to corpus-based methods, by virtue of the fecundity of eighteenth-century composers and their stylistic preference for clearly defined moments of articulation. The present report will give special emphasis to presentations pertaining to this repertory.

The four presentations featured on the opening day set the tone for the entire workshop, implicitly establishing a fundamental tension between two approaches that would unfold more explicitly over the following days. One approach may be construed as the more strictly and systematically method-oriented, similar in its rigour to the exact and social sciences, while the other is a more exploratory, theory-oriented approach, addressing musical corpora more flexibly in order to derive or corroborate theoretical propositions. Martin Rohrmeier (École Polytechnique Fédérale de Lausanne, hereafter EPFL) and Daniel Shanahan (Ohio State University), whose research reflects the first of these two approaches, showed how certain analytical paradigms may be applied to musical information (symbolic representations as well as audio) in order to test theoretical hypotheses. Rohrmeier, largely drawing on methodologies derived from linguistics and information science, showed how nested, ‘context-free’ grammatical structures may be used effectively to model the complexity of musical phrases. Shanahan (‘Measuring the Immeasurable in Music’) offered a reductionist perspective focusing on a limited number of variables, and proposed a fundamental distinction between actively perceived high-level musical features, such as harmonic progressions, and low-level structures that are only implicitly processed by listeners, such as note-to-note transitions. Dmitri Tymoczko (Princeton University) and Yoel Greenberg, on the other hand, took a more flexible approach, integrating quantitative tools with more traditional methods of music research. In his paper ‘What Can We Expect from Corpus Studies?’ Tymoczko argued that painstakingly differentiated rules of harmony, voice leading and the like generally fail to amount to laws without exceptions. He demonstrated how corpus studies are often used to reinforce textbook theories ‘as we know them’, but may sometimes in fact be employed to refute theories instead. Greenberg followed this revisionist path in his investigation of the evolution of sonata form by showing how the analysis of a large body of works may provide new insights into the evolution of common-practice style and musical form in systems-theoretical terms.

Advanced computational tools represent a crucial aspect of modern corpus methods, and nearly the entire second day was devoted to this subject. Fabian Moss (EPFL) extracted the tonal profiles of musical pieces from various stylistic periods based on available MIDI and MusicXML files; this allowed him to highlight the importance of enharmonic spelling in modelling such tonal profiles and to show how even such highly reductionist histogram representations reflect large-scale changes in compositional practice over the centuries. Moss demonstrated how various non-diatonic notes became quite common in different keys during the nineteenth century, indeed an idiomatic part of a tonality, suggesting that we need to reconsider the relationship between the notions of ‘key’ and ‘musical grammar’. He then concluded his presentation with the thought-provoking observation that Bach's Mass in B minor and Liszt's Sonata in B minor are in fact in two different keys. The other presenters in this session discussed further uses of algorithmic methods to explore aspects of musical syntax. Their topics included expressing hierarchical relationships between the background and surface of a piece using graph-based representation of musical structure (Christoph Finkensiep, EPFL); histogram-based tracing of the emergence of the major and minor modes during the baroque period, their consolidation in the classical era and their subsequent dilution from the nineteenth century onward (Daniel Harasim, EPFL); automated analysis of fugues and sonata forms based on pattern recognition (Mathieu Giraud, Université de Lille); and using pitch-class histograms to model musical form (Dror Chawin, Tel Aviv University). Zohar Eitan and Neta Maimon (Tel Aviv University) reported on a series of cognitive experiments investigating cross-modal correspondences between tonal stability and several non-auditory domains (such as visual brightness, physical size and spatial location), showing that participants reliably associate tonally stable notes with brighter, happier and spatially higher visual stimuli. They contemplated some possibilities for furthering their investigation by applying corpus-based methods to repertories combining auditory and non-auditory content (for instance, art songs, opera and film music).

The final two days of the workshop concentrated on eighteenth-century music from a variety of perspectives. Jan Miyake (Oberlin College) and Uri Rom, focusing on Haydn and Mozart respectively, challenged some aspects of James Hepokoski and Warren Darcy's sonata theory. Based on a large sample analysis of Mozart's sonata expositions in various instrumental genres, Rom showed that deviations from the textbook formal trajectory tend to occur in conjunction with longer movements, thus calling into question the applicability of the concept of ‘deformation’ to such constellations. Miyake's investigation into the last movements of Haydn's symphonies underlined the almost omnipresent role played by the recurring main theme, leading her to propose the concept of thematic saturation, instead of the traditional notion of monothematicism, to describe Haydn's techniques.

Corpus studies also prove useful in identifying composer-specific features. Bella Brover-Lubovsky (Jerusalem Academy of Music and Dance), for example, examined Vivaldi's output in its entirety and provided valuable insights into his musical syntax. She demonstrated Vivaldi's use of bass techniques, including the lament bass, the falling-fifths sequence and pedal points, their employment at different positions across musical phrases (Vordersatz, Fortspinnung, prefix and cadence) and their interaction to form periods. Yosef Goldenberg (Jerusalem Academy of Music and Dance) pinpointed various features that are used consistently in Schubert's short piano dances, including the non-rounded binary form and hypermetrical shifts. Markus Neuwirth (EPFL) and Johannes Hentschel (EPFL) presented a statistical survey of cadence types and their corresponding harmonic progressions in Mozart's piano sonatas, showing cadential six-four-chord figures to be the most prominent markers of an imminent perfect authentic cadence, while ii⁶ represents the most frequent pre-dominant harmony. Mozart was also the focus of the presentation by Nathan Martin (University of Michigan). Martin defined four categories of half cadence in Mozart's music and has trained a computer program to detect them, with the aim of relating these half-cadential types to various formal functions. Some success in machine-learning identification of cadences was achieved by Ben Duane (Washington University in St. Louis), whose approach is based on extracting harmonic and melodic cues from music, and by Matan Ben-Asher (Bar-Ilan University), in his work in progress on cadence detection using finite-state machines.

Another benefit of corpus studies is their suitability for investigating generic norms and formal conventions. In her study of eighteenth-century minuets, Olga Sánchez-Kisielewska (University of Chicago) discovered subtle, yet substantial, differences in hypermetre and phrase structure between ballroom minuets and symphonic minuets. Her talk provided an especially rich synthesis of corpus scholarship with theoretical and historical approaches, including some deriving from dance theory. My own paper (Omer Maliniak, Bar-Ilan University) followed up on Greenberg's research, demonstrating the increasing importance of the solo sections, in terms of both thematic substance and proportions, as a catalyst in the emergence of the classical concerto from baroque ritornello form.

The workshop ended with an open discussion regarding the future of corpus studies and computational methods. Participants agreed that the relevance of traditional scholarship notwithstanding, the powerful combination of computers and large bodies of works – enabling researchers to test various hypotheses on a vast amount of music in a short amount of time – will require scholars to adopt and adapt these means in order to find new scholarly pathways. Some speculated that musicology as a discipline will see paradigm shifts and the integration of methods from other, more ‘exact’, fields with the traditional ones, but will first have to improve its methodological standards. Finally, participants highlighted their sense of being part of a community of corpus researchers and expressed their willingness to engage in future collaborations and joint projects.

I would like to acknowledge the cooperation and support of Uri Rom and Yoel Greenberg in the writing of this report.