5.1. Premise and Paraphrase.
5.2. Analyses of Corpora.
5.3. A Sample Corpus.
5.4. Native vs Non-Native Musical Speech.
[5.1.1] “Particular languages are learned by generalizing over utterances that a learner has heard used, while language production and comprehension involve combining or decomposing an utterance into its more basic form-function correspondences (Bybee 1985, 2006, 2010, 2013; Langacker 1988; Barlow and Kemmer 2000; Verhagen 2002; Lieven et. al. 2003; Tomasello 2003; Goldberg 2006; Abbot-Smith, Dittmar, and Tomasello 2007; Alishahi and Stevenson 2008)” (Goldberg 2013, 27).
[5.1.2] We learn language from experiencing its usage in context. We can remember some utterances verbatim or nearly so, and we can also categorize utterances by their similarities. Recent research has found that a two-year-old child may make more than 60% of its utterances from verbatim recyclings of previous utterances (Lieven et. al. 2003). And adult speech and writing has been found to consist of more than 50% “prefabs” (prefabricated expressions; Erman and Warren 2000), meaning learned constructions and collocations whose ensembles of words occur more frequently than could be predicted by the individual words’ separate probabilities. According to Goldberg (2013, 28), “There is a great deal of evidence demonstrating that we retain an impressive amount of item-specific knowledge including relative frequencies of usage.” An improvising musician’s ability to work within the generic patterns of a chosen style while simultaneously avoiding unintended direct quotations from any of thousands of compositions reflects the ongoing interplay between verbatim and schematic memories.
[5.2.1] Early computational research into corpora of written and spoken language uncovered thousands of statistically significant relationships between pairings of words (Sinclair 1991). Some of these pairings had previously been recognized as idioms—word combinations with figurative or metaphorical meanings (e.g., see eye to eye, or sea legs). Thousands of others with more ordinary meanings—“prefabs” like pick and choose, or good friends (Bybee 2006)—had largely escaped notice. In the movie Forrest Gump (Forrest Gump 1994), a brief dockside exchange between a double amputee and the eponymous main character demonstrated the embarrassing difference between knowing and not knowing a common collocation:
Lieutenant Daniel Taylor: I’m here to try out my sea legs.
Forrest Gump: But you ain’t got no legs, Lieutenant Dan.
For non-native speakers of a language, learning high-frequency collocations is a practical necessity if they aspire to native fluency, and the enormity of the task in English is demonstrated by the 250,000 collocations listed in the Oxford Collocations Dictionary (2009). Note that this number far exceeds the number of single words in a typical person’s vocabulary, suggesting that the bulk of linguistic knowledge in memory involves words in combination.
[5.2.2] Listeners’ sensitivity to frequency of occurrence extends beyond just having a sense that a particular construction is rare or common. Listeners also learn the correlations of subsidiary features within a particular construction, which have been termed “collostructions” (Gries and Stefanowitsch 2004). Studies of collocations and collostructions are rare in music, perhaps because access to the kind of digital corpora available in language studies (where one can readily examine millions of words of natural speech or narrative prose) is not yet, or only just becoming, a reality. In a hopeful sign, two recent issues of Music Perception (Temperley and VanHandel 2013, VanHandel and Temperley 2014) examine the state of corpus studies in music.
[5.3.1] In dealing with historical repertories, one can replace an (unattainable) unbiased sampling of an entire historical era with an editorial selection by contemporary musicians of intentionally representative works. For the galant style of eighteenth-century music, such an authentically curated selection was made of Italian solfeggi (pedagogical works for voice and basso continuo) by the editors of the Parisian collection titled Solfèges d’Italie (Levesque and Bèche 1772). Later editions of this collection remained influential in the early years of the Paris Conservatory.
[5.3.2] To begin a pilot study of usage-based knowledge in music, let us first examine a small musical utterance from the Neapolitan composer Gaetano Latilla (see Example 12). Many years later Latilla would teach Thomas Attwood before that young Englishman went on to become a prominent student of Mozart. Latilla’s excerpt comes from his La finta cameriera (Latilla 1737), which in Paris during the 1750s would figure prominently in the War of the Buffoons (Charlton 2012).
[5.3.3] Latilla was replicating a known construction—a conventionalized phrase-type pioneered by composers like Alessandro Stradella (1639–82) or Giovanni Bononcini (1670–1747), one centered around a Romanesca leading into a Do-Re-Mi (Gjerdingen 2007a, chaps. 2 and 6). The question here is whether, given the complexity of Latilla’s exemplar, seemingly inessential details of his usage might be replicated in the later Neapolitan tradition with a frequency high enough to suggest collostructional biases. We are not talking about direct influence in a chain that goes back to Latilla personally, but of a diffuse distribution of commonplaces among the “speakers” of a particular musical “language.” In today’s internet jargon, we might talk of “memes going viral.”
[5.3.4] The nine excerpts of Example 13 (a–i) represent the clearest exemplars of Latilla’s construction in the order in which they occur in the Solfèges d’Italie. Their numbering in that collection is given in the upper left of each exemplar. Numbers 44 and 46 are unattributed, number 64 is by Leonardo Leo (a paragon of Neapolitan style), the next four (92, 189, 192, 206) are all by J. A. Hasse (Naples trained, though more famous for his operas in Dresden), and the last two are by the Neapolitan maestro Pasquale Cafaro, whose music was heard by the young Mozart (Naples, 1770). It should be mentioned that schema theory is neutral as to whether Latilla and others wrote (1) one construction with two or more component schemata, (2) two constructions collocated as a higher-level prefab, or (3) two or more constructions used in a stereotypical pattern of discourse.
[5.3.5] We believe the ten exemplars of Examples 12 and 13a–i are all based on the same construction, though an attempt to define it through a rigid list of necessary features would likely fail given Wittgenstein’s (1953, §66) caution to view a mental category as a “complicated network of similarities, overlapping and crisscrossing.” Following Givón, we view a construction or schema as either a prototype or a central tendency of a class of exemplars associated through similarity. One of many such similarities would be the surprising frequency with which a melodic ornament (circled in dashed lines in Examples 13a–i) is performed at the third stage of the construction, where the bass sounds scale degree ⑥ and the melody sounds ❶. This seemingly inessential ornament is a feature of Latilla’s exemplar (Example 12), of all the exemplars by Hasse and Cafaro, and of both unattributed exemplars. Its high frequency of occurrence suggests that it is a central feature of this construction and not something to be reduced away. There appears to be a strong collostructional bias for an ornament to appear at that location in this construction.
[5.3.6] While that assertion is true as stated, note that the ornament is lacking in the exemplar by Leo (no. 64), and that Leo’s phrase begins with a different melodic shape, moving from ❶ to ❺ and then up to the octave ❶. This exemplar is clearly related to the others, since it begins with the standard bass and concludes with the Prinner bass (④-③-②-① as defined in MGS), something shared with number 227 by Cafaro (note that Latilla’s own exemplar [Example12] concludes with the Prinner melody [❻-❺-❹-❸]). But perhaps Leo’s configuration is sufficiently different for it to qualify as a subsidiary construction with its own set of collocations. There might be multiple subtypes of a common construction, and some of them may stem from different times and locales, resulting in “dialects” or what what Narmour termed “idiostructures” (Narmour 1977, 181).
[5.3.7] The Solfèges d’Italie appear to have a generally chronological ordering, and only the most modern exemplars (i.e., Cafaro) are presented in full ornamental dress. Yet clusters of “overlapping and crisscrossing similarities” are quite apparent, even in this small sample. All of Hasse’s exemplars (nos. 92, 189, 192, 206), for instance, repeat their cadential halves, creating the ABB form of themes common at midcentury, and Latilla’s exemplar (Example 4) shares with Cafaro’s two exemplars (nos. 218, 227) the terminal “High-❷ Drop” (a term from MGS to describe a melodic close where the second scale degree in an upper octave drops a sixth to conclude with scale degrees ❹ and ❸). By contrast, the last of Hasse’s exemplars (no. 206) shares with both of Cafaro’s the melodic start on scale degree ❺ followed by a scalar descent to ❶ (shown by arrows in Example 13). From this pilot study in musical collocation it appears that the collostructional biases of subsidiary patterns may be influenced by the exact subtype of a construction, at least for a high-frequency and well-known construction with many known variants.
[5.3.8] A musical core sample like that of Example 13 aids in recognizing the enormous sensitivity that eighteenth-century court musicians had to preferred collocations and collostructions—after all, those artists were the “native speakers” of this musical idiom. In the dominant Italian style, the greatest stylist was no doubt Mozart, and a fitting capstone to the prior exemplars may be a brief examination of the opening that he wrote for the Adagio of his "Gran Partita," K. 361 (1781; see Example 14).
[5.3.9] Mozart begins the Adagio with a brief, solemn fanfare (m. 1, not shown) and then follows with the Leo subtype of the construction in question (mm. 2–3, not shown, but much like mm. 4–5 shown in Example 14, minus the upper staff). The full construction follows (mm. 4–7 as shown). Had Mozart read the prior discussion of construction subtypes and memorized all the exemplars of Example 13, he could not have been more accurate and knowing in his placement of all the collocated features. The Leo subtype of mm. 2–3 properly received no melodic ornament at the third stage, as would be expected, but the Latilla-Hasse-Cafaro complex of mm. 4–7 did, again following tradition exactly. When the solo oboe enters in m. 4 with scale degree ❺ (upper staff), Mozart provides it the customary stepwise descent to the melodic ornament on ❶. At that point the clarinet enters to begin the ❶–❷–❸ ascent common to most instances of this construction, and in closing, the clarinet executes the High-❷ Drop just as the Hasse-style deceptive cadence prompts a further continuation.
[5.3.10] This description overstates the normality of the phrase because Mozart manages an amazing degree of subordinated complexity. The clarinet’s ascending chromatic appoggiatura
[5.4.1] In response to Mozart’s presentation of this construction and the ensuing movement, the fictional Salieri of the film Amadeus (Amadeus 1984) says, “This was no composition by a performing monkey! This was a music I'd never heard.” While we obviously agree with the first assertion, the real Salieri would clearly have recognized most of this construction from his own compositions and those of many others. The real Salieri might not, however, have heard this level of mastery in the art of combinations, the eighteenth-century’s prized ars combinatoria (Ratner 1970).
[5.4.2] The difference between Mozart, clearly a gifted native speaker of the Italian galant style, and the fine French composer André Grétry, who admitted his limitations in counterpoint even after five years of study as an adult learner in Rome, can be heard in an aria from Grétry’s Les mariages samnites (1776, rev. 1782; see Example 15).
[5.4.3] Grétry’s orchestration is charming and was innovative in its day. Mozart (who knew this opera and wrote a set of keyboard variations, K. 352, on one of its arias) likely recognized in it a useful model for how to expand the size of schematic phrases while simultaneously maintaining rhythmic interest (note how Grétry’s cello plays more than a dozen notes for each of the core bass tones found in Latilla’s construction). But the core melodic tones of Grétry’s exemplar do not match the Italian tradition. Not one of the exemplars in Example 13 matches Grétry’s design. To use a linguistic analogy, Grétry, in spite of his inventiveness and originality, learned the Italian tradition as a second language while a young adult. He did learn parts of the tradition very well (his bass is note-for-note perfect, he inserts a melodic ornament at the third stage, and closes with the High-❷ Drop), but he did not learn all the subtle collostructions and prefabs that would have characterized a native speaker. This is not a judgement based on some absolute standard of musical excellence. It is a judgment supported by detailed knowledge of this Italianate construction as it was replicated in Mozart’s day. It may speak to why Grétry’s ultimate triumphs with comic operas for the French public did not translate into similar success with the Italian public.
[5.4.4] Today, as several scholars begin to explore the computational study of musical corpora, the types of collostructional analyses of individual constructions being pursued in computational linguistics (see Gries and Stefanowitsch 2004) may offer advantages over analyses based on more elementary and global prescriptions of musical syntax (see Gjerdingen 2014 for a further case study).