Roux, J., Giraud, M. & Levé, F., (2026) “Annotating Texture and Imitation Patterns in a Corpus of Slow Movements in Corelli’s Trio Sonatas”, Empirical Musicology Review 20(1), 63-70. doi: https://doi.org/10.18061/emr.v20i1.10018

INTRODUCTION

ARCANGELO Corelli’s oeuvre is almost exclusively made up of instrumental works for string ensembles, including Trio Sonatas (Sonata a Tre) and Concerti grossi, the latter consisting of a dialogue between small groups of soloists and the orchestra. The Italian composer (1653-1713) had a notable influence on his contemporaries and successors (Rothstein, 2006; Sanna, 2013).

As an outstanding violinist and a renowned composer, Corelli was protected by powerful people (Pincherle, 1954). In Rome, Corelli published five sets of 12 Trio Sonatas, for two violins and a basso continuo, each sonata having about 4 movements: the Church Sonatas Op. 1 (1681) and Op. 3 (1689), the Chamber Sonatas Op. 2 (1685) and Op. 4 (1694), and a mixed Op. 5, including the variation on La Folia (1700). An additional collection was published posthumously (WoO 5-10, 1714). In the Trio Sonatas, the violins are equal in their range and their melodic role. Crossings between them are frequent. The continuo may be played by two instrumentalists, one improvising the chordal realization (for example, on the harpsichord, the organ, or the lute), the other playing the continuo melody.

Rameau (1726) analyzed some Corellian cadences in his Nouveau système, which led, among other things, to the theorization of the cadential 6/4. Bergé (2015) discusses how cadences structure Corelli music. Most of them mark transitions between phrases, and sometimes are recurring elements across movements (Libby, 1973). Wintle (2017b states that "Corelli’s trio-sonata textures have been defined in terms of the polarization of two intertwining violins, moving predominantly in parallel thirds, which are set apart from their supporting bass line” (p. 33). Wintle (2017a also analyzes the Corelli sonatas across their rhythms and their modulations, and discusses structure through repetitions and expansions/extensions of tonal and "large rhythmic units" (p. 56). He further states: "Corelli’s oeuvre is founded upon a fairly limited number of musical figures, or models, which are capable of sustaining a considerable variety of modes of presentation" (p. 32).

Here we analyze formal structures, texture, and imitation patterns, trying to describe such Baroque writing techniques. The imitation patterns are repeated, possibly transposed, and possibly used in melodic sequences. We release a dataset on all the 38 slow movements of the 24 Corelli’s Church Sonatas from Op. 1 and Op. 3. These slow movements are relatively short, allowing in-depth analyses, and showcase a variety of textures and writing techniques with a focus on patterns. They already account for 954 measures, making this sub-corpus a suitable option for an initial modeling of textures. We began with .musicxml files released by Hentschel et al. (2021) [1], built by curating data from Craig Sapp’s KernScores [2] and the Pepusch (1740) edition of Corelli’s Trio Sonatas Op. 1 and Op. 3, available from the International Music Score Library Project (IMSLP) [3]. We complement the annotations from Hentschel et al. (2021) with new annotations on texture and writing techniques. The fact that, including us, three different groups have already worked on and improved this material shows the advantage of such open-data research.

Modeling texture and imitations is less common than modeling harmonic functions and is open to debate. With our formalization, we aim to limit certain biases, for example, those introduced by inconsistent or overly intuitive annotation. However, we acknowledge that any analysis of musical texture, a domain that is not yet well established, is inevitably shaped by our perspectives and interpretative choices, which we consider an inherent and acceptable part of the process. An example of this approach can be seen in the Grave of the first Trio Sonata from Op. 1 in F major (Figures 1 and 2).

Figure 1. Analysis of the first movement (Grave) of the Trio Sonata in F major, Op. 1 No. 1, annotations as visualized on Dezrann : https://www.dezrann.net/~/corelli-trio-sonatas/op1.1.1. Patterns are annotated (a, b, c | x [imitation]), and further labels are on phrase textures (Homorhythmy [H], Imitation with suspensions [ID]), pattern textures (homorhythmy [h], imitation [i]), rhythm density (from whole note [1] to sixteenth note [16]), melodic sequences (number of occurrences, direction, diatonic interval), o↗, o↘ (octave jumps) and on cadences (Perfect Authentic Cadence [PAC])

Figure 2. Analysis of the first movement (Grave) of the Trio Sonata in F major, Op. 1 No. 1, using the same notations as in Figure 1.

The Grave contains five phrases, A, B, C, D, D, with the last phrase D being (almost) repeated. Phrase B modulates to C major and phrase D modulates to D minor. Phrases A and C are mostly homorhythmic (H), while phrases B and D have parts with imitation with suspensions (ID). The second phrase B combines some transformed material from A (in homorhythmy) with an imitation of the pattern x (3 occurrences in each voice) that yields a 7-6 suspension in bar 4. The annotations further specify which voices are involved and how they interact. In the first bar of the Grave, the two violins are in homorhythmy (h12). The imitation in B is annotated i[1][23] as the first violin (voice 1) begins, and then the second violin and continuo (voices 2, 3) continues. In the third bar, during the imitations, our interpreted rhythmic density is annotated 8/16, as an “eighth note / sixteenth note” rhythmic pattern that dominates this section. The thematic patterns a, b, and c are exposed in phrase A (and are reused and transformed in later phrases). A short imitation pattern x is repeated and transposed during melodic sequences. The melodic/harmonic sequence in B is annotated as 3↘22, that is three occurrences, descending each by a second (either major or minor).

METHODS

To build the dataset, we first segment and label phrases according to their similarity, using phrase boundaries and cadences from Hentschel et al. (2021) for their delimitation. The movements contain between 2 and 11 phrases (5.26 on average).

The central addition of this dataset is texture and patterns. These annotations take into account both the perception of the musical streams and the written elements that are predominant in the analysis of the score. We identify patterns as musical elements repeated at least twice in the movement, which may arise from a paradigmatic analysis (Ruwet, 1972). Imitation patterns are usually short (3-5 notes), located inside melodic sequences. Patterns may occur in overlapping textural zones. Outside of imitations, one finds thematic patterns in homorhythmy. The same pattern can also be involved in both the end of an imitation and the beginning of the following homorhythmy. The patterns are mostly independent, except in a very few cases, which are too limited in number to be explicitly encoded in the annotation syntax.

Each phrase is thus annotated with its texture (Table 1), either imitation with or without suspension, or mostly homorhythmic (capital letters, H, I, ID). What is considered “mostly” homorhythmic is subject to interpretation: Typically, at least 80% of the beats are synchronized between the involved voices. Some passages may still include slight variations, such as the sixteenth notes in the second bar of Violin II on the Figure 1. In the case where a phrase contains two textures without a clear predominance of one of them, we indicate two textures on a segment, as in the case of phrases B and D in Figure 1, which are respectively annotated H-ID and ID-H, to render the order in which the textures occur.

We then delimited the homorhythmy and imitation zones (lowercase letters, h/i/a, Table 2), indicating the involved voices (see h12 on Figure 1). We start imitation zones from the beginning of the first pattern and stop them at the end of the last one. Homorhythmy is annotated even when the homorhythmy is not “perfect”, as in the first measures of the Grave. Some homorhythmies are rather annotated as parallelism (p). Note that zones may overlap. Moreover, some homorhythmic zones (h) are annotated in non H phrases. Crossed imitations with exchange are rare (1.7.2 and 1.12.1)

Table 1. Annotations on structure and on textures

Table 2. Annotations on patterns, melodic sequences, and octave jumps. Note that there are never more than 5 thematic and 8 imitation patterns in a single movement of our corpus.

We complete the annotation of patterns by non-overlapping annotations for rhythmic densities, labeling the most represented rhythmic figures in each zone, the hemiolas, the melodic sequences, and octave jumps (Table 2). The dataset is completed with harmonic data from Hentschel et al. (2021) (Table 3).

Table 3. Annotations on structure and harmony from Hentschel et al. (2021)

DATA AVAILABILITY

The dataset contains over 7,000 labels. The annotations were proposed by the first author and curated by all authors. We also provide bar-level synchronization, achieved on the Dezrann platform (Garczynski, 2022), for a recording by the London Baroque orchestra directed by Charles Medlam (Harmonia Mundi, 2007). Currently, we are unaware of any complete open-source recordings of these sonatas. However, we intend to publish synchronization data for them when they become available.

The dataset, comprising scores, annotations, and synchronization, is accessible via open licenses (Open Database License, Database Contents License) in .json format on a git repository accessible from algomus.fr/data. It is also accessible on the Dezrann platform (dezrann.net/explore/corelli-trio-sonatas), featuring synchronized playback with the recording.

DESCRIPTIVE STATISTICS

Annotating textures helps us to understand the construction of Corelli’s sonatas. The instruments play on average 41% in homorhythmy (including 8% in parallelism) and 43% in imitation (Figure 3). The vast majority of imitations occur between the two violins and thus, the continuo rarely plays in imitation (less than 5%). Conversely, the continuo frequently plays in homorhythmy with the violins, though a bit less with the first violin.

Figure 3. Homorhythmy, parallelism, and imitation across the three parts

Table 4. Melodic sequences according to their direction, number of repetitions, and intervals

Most of the melodic sequences have two occurrences (ascending or descending ones) or three occurrences (mostly descending ones). Most of the intervals in those melodic sequences are descending minor and major seconds (Table 4). Octave jumps are mostly (69%) located in the bass, with 64% of them being descending ones, often just before the cadence, as in bars 2 and 13 of the Grave. Conversely, 79% of the octave jumps on the violins are ascending (Table 5). The majority of hemiolas are positioned at the end of the movements (Figure 4), creating a slowing-down effect.

Table 5. Ascending (o↗) and descending (o↘) octave jumps on the three parts

Figure 4. Localization of the hemiolas

PERSPECTIVES

Overall, this dataset offers a more abstract and high-level view of musical structure than Hentschel et al. (2021), especially by annotating common textures in the Baroque style, imitation patterns, and melodic sequences. Defining texture is complex and there is no established consensus on how to annotate it. Here, we proposed a consistent method to describe these textures with greater precision. We believe that this annotation data can be valuable for systematic musicology and computer music, benefiting people working not only on Baroque music but also on other styles. Indeed, imitation textures and other high-level musical structures occur across various genres, making this dataset a useful resource for projects in music analysis and music generation. For instance, what does it mean to have a slow movement where the second phrase features imitation with a specific pattern and ends on a half cadence? Being able to model this can support both human and automated analysis, as well as generative approaches, whether through procedural algorithms or machine learning.

More broadly, this initiative represents a step toward formally modeling and discussing musical texture, providing a framework for understanding this essential aspect of music. However, as the dataset is focused on these slow movements, which are relatively homogeneous, the rhythmic variety and contrast may be here limited. Further studies could target fast movements in Corelli's sonatas and other works to model textures more comprehensively.

NOTES

[1] https://dcmlab.github.io/corelli/

[2] http://kern.humdrum.org/search?s=t&keyword=Corelli

[3] https://imslp.org/wiki/Category:Corelli,_Arcangelo

REFERENCES

Allsop, P. (1992). The Italian 'Trio' Sonata: From Its Origins until Corelli. Oxford Monographs on Music.

Bergé, P., & Neuwirth, M. (2015). What is a Cadence?: Theoretical and analytical perspectives on cadences in the classical repertoire. Leuven University Press. https://doi.org/10.2307/j.ctt14jxt45

Garczynski, L., Giraud, M., Leguy, E., & Rigaux, P. (2022). Modeling and editing cross-modal synchronization on a label web canvas. In Music encoding conference (MEC 2022). https://doi.org/10.17613/krxd-pq66

Hentschel, J., Moss, F. C., Neuwirth, M., & Rohrmeier, M. (2021). A semi-automated workflow paradigm for the distributed creation and curation of expert annotations. In International Society for Music Information Retrieval Conference (ISMIR 2021). https://doi.org/10.5281/zenodo.5624417

Libby, D. (1973). Interrelationships in Corelli. Journal of the American Musicological Society, 26(2), 263-287. https://doi.org/10.2307/830667

Pepusch, J.C. (1740). The score of the four setts of sonatas compos'd by Arcangelo Corelli, for two violins & a bass. Ed. Benjamin Cooke, London, 1st éd. Thomas Baker Collection at the University of Western Ontario Music Library https://opac.rism.info/rism/Record/rism990011184

Pincherle, M. (1954). Corelli et son temps. Edition Le Bon Plaisir / Plon.

Rameau, J. P. (1726). Nouveau système de musique théorique. Ballard. Chap. 23, 94-106.

Rothstein, W. (2006). Transformations of Cadential Formulae in the Music of Corelli and his Successors, Third International Schenker Symposium.

Ruwet, N. (1972). Langage, musique, poésie, Paris, Éditions du Seuil, 100-134.

Sanna, A. (2013). Contrapuntally crafted, harmonically eloquent. Corelli's sonatas and the compositional process in the late 17th century. Basler Jahrbuch für Historische Musikpraxis, 37, 73-88.

Wintle, C. (2017a). Corelli's Rhythmic Models: Dance Movements in the Seventeenth and Eighteenth Centuries. Intégral, 31, 51-61.

Wintle, C. (2017b). Corelli's Tonal Models: The Trio Sonata Op. 3, No. 1. Intégral, 31, 31-49.