Corpus

Timeline of Composers by Birth Year

The timeline graph represents the lifetime (shown as green bar) of each of the twenty composers featured in the corpus, listed chronologically by birth year (Leoš Janáček, 1854, to Benjamin Britten, 1913), representing a total time range of 1854 to 1992 (shown with blue stars). Although the corpus includes pieces representative of each of the creative periods of the composers' lifetime up to 1969, works were selected with the focus period of 1900-1950 (shown with double-sided orange arrow). The graph allows the viewer to identify the composers' creative periods that fall within this focus period. For example, the focus period corresponds to Claude Debussy's (1882-1918) middle to late period, while it corresponds to Benjamin Britten's (1913-1976) early to middle period. The focus period also captures some composers' entire creative output, such as that of George Gershwin (1898-1937) and Arthur Honegger (1892-1955). The alignment of composers' lifetime and their creative periods also suggests possible encounters between composers as well as familiarity with other composers' output.

Birth Region or Country

The word cloud represents the corpus from the perspective of composers' birth countries. France is the most common birth country, shared by four of the twenty composers (Claude Debussy, Maurice Ravel, Edgard Varèse, and Olivier Messiaen).

Death Region or Country

When considering the corpus by composers' death countries, we see a clear shift to the United States as the most shared country. This shift is in part related to the exile of many composers during World War II, and includes European-born composers Béla Bartók, Arnold Schoenberg, Igor Stravinsky, and Edgard Varèse. When considering the American-born composers Aaron Copland, Charles Ives, and George Gershwin, this corresponds to seven out of 20 composers (35%) with the United States as death country. The composers map provides another visual representation of this shift as well as other intercontinental movements.

Works by Composer

The word cloud presents corpus source works by composer. We can observe that the corpus is most strongly represented by the works of the Czech composer Bohuslav Martinů and the German composer Paul Hindemith, and more weakly represented by the works of the American composer George Gershwin. While this may be related to some bias on the part of Suter (1980), it is also related to his sampling method, which involved taking into account the compositional output of each composer in terms of time period and productivity. Thus, more works were sampled for composers with a larger body of works than for composer with a smaller one. For a list of the number of source works by composer, visit the Composers page.

Works vs. Examples

This graph provides a percentage comparison of works surveyed against works sampled and extracted examples. The percentages are based on the total number of works and examples in the corpus. As in the source works by composer word cloud, Martinů and Hindemith have the highest percentages in all categories, although the percentage of extracted examples for Martinů is lower than the number of works surveyed and sampled. The graph also shows that the percentage of extracted examples is much higher than the surveyed and sampled works for the French composer Olivier Messiaen. This is the result of the relatively large number of examples (20) extracted from Turangalîla, a Symphony for Solo Piano, Ondes Martenot and Orchestra (Mes-22). A similar pattern can be observed for extracted examples from works by the German composer Alban Berg and the American composer Charles Ives. These differences may also be related to those composers' more frequent use of polyrhythm as a compositional technique as compared to the other corpus composers. For a list of the number of works and examples by composer, visit the Composers page. To identify examples by work, visit the Browse page.

Composition Year by Decade

The graph and accompanying table present the corpus works by year of composition. The graph shows a slightly left-skewed distribution of works, with a peak in the 1920s (108 works) and a larger number of works in the decades after (216 works as compared with 126 works before 1920). To sort or select works by year of composition, visit the Browse page. You can also do a quick search by decade on the Home page.

Genre Percentage Distribution

This pie chart presents the proportional representation by source works genres. The most represented categories are orchestral and instrumental works, each of which represent about a third of the corpus. The last third is shared between vocal (12.8%) and stage (13.1%) works. To identify examples by genre, visit the Browse page.

Sub-genre Percentage Distribution

This pie chart shows how each genre is subdivided into sub-genres in the corpus. For example, programmatic works represent a smaller proportion within orchestral works (7%) than symphonic (16.4%) and concerto (14.6%) works. While vocal and stage works are more evenly partitioned, instrumental works are represented by more chamber (24.3%) than keyboard (11.7%) works, whith chamber works representing the largest sub-genre featured in the corpus.To identify examples by sub-genre, visit the Browse page.

Premiere Locations

The location of the first public performance for each source work in the corpus was investigated. Out of 447 venues identified, only 266 still exist. From this research, several countries and cities emerge as centers for new music. At the top, we see United States and France as major centers, with Paris and New York being the premieres cities for a combined total of 140 of the 450 works sampled. Also prominent are Basel, Geneva, and Lausanne counting 29 of the 45 works premiered in Switzerland. Other important countries include Germany, Czechnia, and Austria, with Vienna, Prague, and Brno being among the leading cities where sampled works were premiered.

Ensembles

This word cloud presents the ensembles that were identified for the corpus source works' premieres. Two ensembles are featured prominently, i.e., the Boston Symphony and the Basel Orchestra, which can be explained in part by the fact that both of these organizations were important for the commission of new works in the surveyed period. Also prominently represented are the New York Philarmonic Orchestra and the Philadelphia Orchestra.

Conductors

This word cloud represents the conductors identified for source works' premieres. The relation with specific orchestra is clearly represented, with Paul Sacher (the founder and director of the Basel Orchestra from 1926 to 1987) and Sergei Koussevitsky (the director of the Boston Symphony Orchestra from 1924 to 1949). Also notable is Ernest Ansermet, a Swiss conductor who worked with Diaghilev's Ballet Russes from 1915 to 1923, and founded his own orchestra in 1918, the Orchestre de la Suisse Romande, with which he toured Europe and the United States. Premiered works by Ansermet and the Orchestra de la Suisse Romande include Igor Stravinsky's Symphonic Poem, Pesnya solov′ya (Song of the Nightingale), in Geneva, Switzerland, and several works by the Swiss composer Frank Martin (1890-1974).

Soloists

This word cloud presents soloists involved in the premiere of corpus source works. Of note is the premiere of works by composers, including Paul Hindemith, Olivier Messiaen, and Béla Bartók. Also prominent is the role of Yvonne Loriod (1924-2010), a French pianist, teacher and composer, as well as Messiaen's second wife, who premiered eight of Messiaen's piano works featured in the corpus. Other soloists featured are Peter Pears (1910-1986), an English tenor who premiered several corpus works by Benjamin Britten, and Maurits Frank (1982-1959), a leading cellist of his time, who was featured in several of Hindemith's chamber works as well as Anton Webern's Sechs Bagatellen (Six Bagatellen) for String Quartet, op. 9 (Web-12).

Examples by Composers

The distribution of corpus examples by composer is similar to the distribution of source works by composer, although there are some differences. One prominent difference is the number of examples by the French composer Olivier Messian (68 examples extracted from 21 works), which is in part due to the larger number of examples extracted from Turangalîla (Mes-22; 20 examples). Other composers that come to the fore in terms of examples as compared with works distribution are Charles Ives (47 examples extracted from 23 works) and Arnold Schoenberg (45 examples extracted from 25 works).

Length of Examples by Composer

Examples in the corpus differ in terms of length measured in seconds, from 1 to 122 seconds. In this representation, boxes represent the 2nd and 3rd quartiles of examples' durations. The red number above each plot gives the mean example duration for each composer, and the number in parentheses after the composer name is the number of examples for that composer. Examples by Maurice Ravel (1875-1937) and Olivier Messiaen (1908-1992) are more varied in length, while those by Edgard Varèse (1883-1965) and George Gershwin (1898-1937) are the least varied, with median lengths of 12 and 7 seconds, the latter representing the shortest median duration. There are also a few variants, the longest being an example extracted from Paul Hindemith's Sonata for Four Horns (Hin-97, T673) and another extracted from Olivier Messiaen's Les Corps glorieux (Glorious Bodies) for Organ (Mes-16, T732), both of a duration of 122 seconds.

Full Unit Start Tempo

The word cloud reveals that the start tempo expressions of the source work full units (i.e., in multimovement works or collections, the specific movement or piece from which the examples was extracted) mostly fall into a few common categories, i.e., Allegro, Moderato, Andante, Poco allegro, and Lebhaft (Lively) or are specific to a given work. There are a few tempo expressions that appear more than once, such as Molto moderato et Lento Although most of the tempo expressions are in Italian, there are several in German and French (e.g., "Modéré" vs. "Moderato" and "Lent" vs. "Lento), which tend to be preferred by composers of those nationalities.

Example Start Tempo

The tempo expressions found at the beginning of the examples also fall into a few categories that are shared with those of full units, i.e. Allegro, Moderato, Allegro moderato, and Poco allegro, but wihtout other tempo expressions coming to the fore.

Start Time Signatures by Type

For the purpose of this study, time signatures were categorized into types. The bar graph uses the traditional categories of general compound and odd types that do not differentiate between duple or triple subdivision of the measures at the notated beat level (e.g., 6/8 vs. 9/8 and 5/4 vs. 7/4), and more specific simple meters (duple, triple, and quadruple). Two more categories are used here, i.e., polymetric and missing, the latter combining unmeatered examples and examples where a time signture has not been identifioed due to unavailable sources. For this graph, it appears that simple meters are more common than compound, odd, or polymetric time signatures, with simple quadruple being most represented at the beginnng of full units and simple triple most represented at the beginning of polyrhythmic examples.

Time Signature Categories (RDS)

Because the traditional categories show a strong bias toward simple meters, we opted for a more fine-grained and less biased categorization, which is represented in the two pie charts. The first pie chart provides the start time signature types of the reserve dataset examples, a stratified sample of 80 examples (4 examples for each of the 20 composers in the corpus) used for the first two experiments on listeners' perceived emotions. The second pie chart presents the start time signature types for the testing dataset examples, i.e., the remaining examples in the corpus. It can be observed that RDS's time signature types are generally representative of the time signature types features in TDS. Another general observation is that changing time signatures represent a larger proportion of examples (14.1% anf 15.3%) than polymetric and polydurational combined (6% and 6.7%), which is unexpected. Thus, it would appear that time signatures are not the principal means by which composers achieve polyrhythmic structures in the surveyed historical period. This categorization also affords comparison within the traditional categories. For example, in TDS, compound duple, which is featured in 50 examples, is much more common than compound triple (8 examples) or compound quadruple (4 examples). Similarly, it would appear that uneven duple meters (13 examples in RDS and TDS combined) are more common than uneven triple meters (7 examples). In contrast, the TDS examples that feature simple meters are more evenly distributed (duple = 20%, triple = 22%, and quadruple = 17%).

Time Signature Categories (TDS)

Polymetric Categories (FDS)

This pie chart presents all the combinations of time signatures used for polymetrioc and polydurational examples in the full dataset (FDS) of this corpus. Overall, composers show much creativity in their use of combined time signatures. Nevertheless, two combinations emerge as preferred polymeters, i.e., 3/4 against 9/8 (9 examples) and 2/4 against 6/8 (8 examples), together representing 41% of polymeters. Interestingly, these two combinations present 3:2 at different levels. The first consists of a 3:2 relationship at the subdivision level, while the second consists of a 3:2 relationship at the beat level, although these represent theoretical possibilities rather than actualizations, which can only be ascertain by close study of the examples.