1 Introduction

1.1 Sociolinguistic Background

Because of documentary and literary reasons, the language of Florence is probably the most studied among the Italian dialects. Its historical development, though, has not been as linear as one might think.¹ Indeed, as is the case for many other dialects, it begins to be widely written, also for literary works, during the Late Middle Ages, but—as is not the case for many other dialects—it suddenly becomes extremely popular, thanks to famous poets that used it and spread it all around the Peninsula in the 13^th century.²

It goes without saying that the 14^th century permanently sanctions the supremacy of the Florentine language. The ‘Three Crowns’—Dante, Petrarch, and Boccaccio—ennoble it by writing masterpieces the caliber of the Comedìa, the Rerum Vulgarium Fragmenta, and the Decameron.

Thus in Florence, in the subsequent centuries, the spoken language is constantly evolving over time, as is expected to be. However, people from outside Florence, who increasingly need a common code to communicate, take the language of the 14^th century as a reference point. Thus, 14F acquires a greater value than other Italian dialects both because of the literary importance of the texts of Dante, Petrarch, and Boccaccio, and because—unlike other contemporary dialects—it is a written model, which could be studied and learned.

From the end of the Quattrocento and throughout the Cinquecento, some humanists begin to recommend taking as a linguistic model the 14F (often called ‘volgar lingua’), disregarding any further development occurred at the spoken level. The first two grammar books are the Regole grammaticali della volgar lingua by Giovan Francesco Fortunio (1516) and the Prose della volgar lingua by Pietro Bembo (1525)—the latter being by far the most important and most widespread of its kind.³

Surprisingly enough, in the sixteenth century, people from outside Florence are much more prone to learn 14F than Florentine people themselves. Indeed, the foreigners study 14F as a completely new language, without any concern for the fact that it is a ‘dead’ language, which dates back two hundred years. On the contrary, the Florentine people hardly accept to use a language different from the one they speak and write in daily life. Unfortunately for them, the more time passes, the more 14F is perceived as the language of high society: people from Florence are increasingly required to use it, too, because this is how they are expected to communicate in cultured and educated milieus.

Indeed, in the whole Cinquecento, we notice in Florentine texts a tension between the will to keep the contemporary language (= the 16F), and the need to use the 14F to communicate with people from outside Florence. In this context, it is therefore of great interest to analyze the historical evolution of the Florentine language throughout the entire 16^th century.

After performing a correspondence analysis and a correspondence regression on Michelangelo’s entire epistolary corpus (about 500 letters), I verified an evolution over time in his use of the language, and I provided a historical explanation to the outcomes of the statistical tests.

1.2 Michelangelo’s Language: An Open Question

In this paper, I focus on Michelangelo’s epistolary language. On the one side, I have chosen to analyze only letters (leaving aside the many poems written by the sculptor) because the current linguistic studies increasingly show the need to focus mostly on practical texts instead of literary works.⁴ Indeed, practical texts are the best choice for linguistic analyses, because they do not aim to be artistic, and frequently belong to authors without any specific literary education ((Serianni 2007): 13).

In recent years, the relevance of private correspondence has been quickly perceived by scholars (cf. e.g., (Magro 2014): 106). Letters provide a wealth of precious information for linguists, both because they often carry a date and because their language is frequently close to that of ordinary speech, thus offering access to useful data.⁵ As is shown, for example, in (Merja Kytö 2010): 17, letters are included in the group of speech-related genres and listed among the speech-like typologies. Thus, despite in epistolary writing the interaction takes place asymmetrically over time, communication is similar to that of the oral speech.⁶

Specifically, this research is targeted towards Michelangelo because he can be considered one of those “intermediate individuals, neither erudite not uneducated people,”⁷ who are nowadays drawing the attention of scholars. As is well known, since the appearance of the notion of semi-educated writers ((Bruni 1978), (Bruni 1984)), scholars increasingly discussed the topic, and today there is a strong tendency to consider the writers’ level of education as a continuous, rather than as a set of discrete variables ((Fresu 2004), (Librandi 2004), (Bianconi 2013)).

Finally, it is also of interest to study Michelangelo’s language because it fits in a complex and much debated epistemological framework, that of the language of arts and artists. Starting with (Folena 1951) and (Folena 1957), scientific studies on this topic have multiplied, and have involved prominent scholars, from (Barocchi 1984) to (Nencioni 1995).⁸

Previous scholars have argued that Michelangelo’s epistolary language constitutes a representative example of 16F, and that it does not make use of most of the features that characterize the language of the Three Crowns.⁹ Persuasive as this may seem, I suggest that this assumption can be challenged.¹⁰

The language of Michelangelo’s letters testifies to an interesting tension between the contemporary linguistic usage typical of a sixteenth-century man of Florence, and the Old Florentine literary language prescribed by the grammarians. I present in Section 2.2 the results of an investigation conducted so as to determine the extent to which Michelangelo used 14F and 16F. Indeed, from my findings, it would seem that the artist was more aware of the Old Florentine linguistic system than was initially assumed, as it appears that, starting from 1530, he was not loth to borrow from it.

2 Methodological Framework

2.1 Limits and Constraints

Before I go any further, I cannot pass over in silence that the boundaries between 14F and 16F are less clear than suggested above. As is well known, some of the so-called fourteenth-century linguistic phenomena had already occurred by the end of the Duecento and the beginning of the Trecento, but the majority of them only appeared in its second half, and became more stable during the following century.¹¹ Moreover, it is not even clear when exactly those phenomena started to fade. It is probable that some phenomena were still in use in the first half of the sixteenth century, while others had spontaneously evolved, and others still suddenly found themselves in competition with the fourteenth-century linguistic system, which—at some point—replaced them. Accordingly, the labels 14F and 16F do not reflect a clear chronological distinction, and each phenomenon should be discussed and evaluated on a case by case basis.

Another issue is that I focus only on the diachronic variable, while I do not take into account the diaphasic variation. Obviously, for a more comprehensive approach, I should distinguish between letters sent e.g. to subordinates or relatives, and letters sent to the pope or to noblemen. The contents of the message hardly are the same, and the overall tone and style can vary significantly from letter to letter. However, because of the high number of documents taken into account, considering uniquely the diachronic variable can lead to interesting results, too.

Two other limits are somehow inherent to such research. First, I analyze only one writer, while—for a comprehensive study of the variation of the Florentine language in the 16^th century—many epistolary corpora, written by different authors, should be compared. And second, the open debate about the possibility of determining (and to what extent) new information about a spoken language from the analysis of its graphic representation, dates back to the creation of the word scripta itself ((Remacle 1948)). However, as (Arcangeli 2011): 10 (translation mine) notes: “if we are willing to formulate some hypothesis on the state of [a language] between the Middle Ages and the Renaissance, […] we are necessarily forced to base our conjectures on written texts.” Indeed, historical linguistics “is not a second-best solution by inevitable necessity, but just the best solution in those areas of study for which oral records are not available, especially when studying long-term developments of language variation and change” ((Natalie Schilling 2012): 64).

Then, strictly speaking, I am analyzing only the scripta—not the language—of Michelangelo: but analyzing the scripta is the only way to get information about his language.

2.2 Results

To collect the corpus, I copy-pasted the texts from (“Memofonte,” n.d.) to thirteen .txt files, split into time intervals, from 1495 to 1564.¹² At the end of this first step, each file contained all the letters written by Michelangelo over a range of five years.¹³

Then, I deleted all the special characters “.,;:!?’·” and I split the texts, one word per line. At the end of this step, I put every document (= every time interval) into vectors.¹⁴ Subsequently, based on the current bibliography,¹⁵ I selected the major features that differentiate 14F and 16F. For every feature, I identified the golden forms (i.e., 14F forms) and the silver forms (i.e., 16F forms). I display here all of them:

• CruoG = A tonic Latin ŏ, preceded by consonant plus r, becomes uo in 14F (ex. Lat. prŏba > pruova)

• CruoS = A tonic Latin ŏ, preceded by consonant plus r, becomes o in 16F (ex. Lat. prŏba > prova)

• CrieG = A tonic Latin ě, preceded by consonant plus r, becomes ie in 14F (ex. Lat. brĕvis > brieve)

• CrieS = A tonic Latin ě, preceded by consonant plus r, becomes e in 16F (ex. Lat. brĕvis > breve)

• schiVG = Before vowel, /skj/ remains /skj/ (written <schi>) in 14F (ex. Lat. sclavus > schiavo after a passage cl > chi in Medieval Latin)

• schiVS = Before vowel, /skj/ becomes /stj/ (written <sti>) in 16F (ex. Lat. sclavus > stiavo)

• lliG = A plural noun or adjective, ending in -lli in Latin, ends in -lli in 14F (ex. Lat. caballus, pl. caballi > cavallo, pl. cavalli)

• lliS = A plural noun or adjective, ending in -lli in Latin, ends in -gli in 16F (ex. Lat. caballus, pl. caballi > cavallo, pl. cavagli)

• artG = The singular masculine definite article is il in 14F

• artS = The singular masculine definite article is el in 16F

• prIVamoG = The indicative present IV p. ends in -(i)amo in 14F (ex. noi scriviamo)

• prIVamoS = The indicative present IV p. ends in -(i)ano or -emo in 16F (ex. noi scriviano, noi scrivemo)

• prVIanoG = The indicative present VI p. ends in -ano in 14F (ex. loro parlano)

• prVIanoS = The indicative present VI p. ends in -ono or -eno in 16F (ex. loro parlono, loro parleno)

• impfVIvanoG = The indicative imperfect VI p. ends in -vano in 14F (ex. loro scrivevano, loro mangiavano)

• impfVIvanoS = The indicative imperfect VI p. ends in -vono or -veno in 16F (ex. loro scriveveno, loro mangiavono)

• futrG = The endings of the indicative future have one intervocalic r in some verbs in 14F (ex. scriverò, scriverai, scriverà...)

• futrS = The endings of the indicative future have two intervocalic r in some verbs in 16F (ex. scriverrò, scriverrai, scriverrà...)

• congG = The subjunctive present ends in -a (I, II and III p.) and -ano (VI p.) in 14F (ex. che tu abbia, che lui voglia, che loro debbano...)

• congS = The subjunctive present ends in -i (I, II and III p.) and -ino (VI p.) in 16F (ex. che tu abbi, che lui vogli, che loro debbino...)

• condG = The endings of the conditional present have one intervocalic r in some verbs in 14F (ex. scriverei, scriveresti, scriverebbe...)

• condS = The endings of the conditional present have two intervocalic r in some verbs in 16F (ex. scriverrei, scriverresti, scriverrebbe...)

• trG = Lat. de ĭntro, de rĕtro become dentro, dietro in 14F

• trS = Lat. de ĭntro, de rĕtro become drento, dreto in 16F

• senzaG = Lat. absĕntiā becomes senza in 14F

• senzaS = Lat. absĕntiā becomes sanza in 16F

• ultimG = Lat. ultĭmus, -a, -um becomes ultim- in 14F

• ultimS = Lat. ultĭmus, -a, -um becomes utim- in 16F

Clearly, here I am only talking of formal variation, significantly different from conceptual variation. The latter refers to the authors’ choice between the use of a word (for instance, ‘oak’) and, e.g., the use of an hyperonymous (‘tree’), while the former only concerns the linguistic variation of the same term—such as, in sixteenth-century Florence, the choice between the forms ‘senza’ and ‘sanza’, both meaning ‘without’. Within this approach, “the downside is that formal variation is only one aspect of a much broader reality, but it is an aspect we claim is worth isolating” ((Dirk Speelman 2003): 319).

After listing all those features, I ran the corresponding queries all over the thirteen vectors, so to find the total number of occurrences of each feature for each time interval, and then I manually put the outcomes in a single .csv file. I underline that for running the queries, I had to choose between two options. Sometimes, I could search for the exact match. That was the easiest way. So, for example, to find the occurrences of senzaG and senzaS I could use those scripts: senzaG <- '\\bsenza\\b' and senzaS <- '\\bsanza\\b', and thereafter, I calculated the total number of occurrences with the function ‘length’. So, for counting the occurrences of the silver form ‘sanza’ in the letters written between 1495 and 1499, I used the script: length(conc_re(senzaS, a, as_text = TRUE)$match). After that, I recorded the numerical outcome in a separated file.

Sometimes, however, I could not search for the exact match. In this case, for every occurrence, I had to manually check whether the outcome was correct. I did it with the function ‘View’, like this: View(conc_re(CrieG, a, as_text = TRUE)). Below, I explain with a few examples why, under certain circumstances, it was impossible to search for the exact match in a completely automated way.

It can happen that a similar visual outcome represents different grammatical rules: for instance, a software cannot make the distinction between the form ‘scriviano’ (that belongs to the prIVamoS group) and the form ‘pregano’ (that belongs to the prVIanoG group), because—graphically—the two words have the same ending -ano (stressed in the first case, unstressed in the second case). The solution that I found, was to write the same script for the two features—'(ano\\b)'—, and then disambiguate them on a case by case basis, depending on the context.

Sometimes—as in the case of the CrieG, CrieS, CruoG, CruoS groups—the rule applies only to words derived from Latin short vowels (for instance, from Lat. prĕcari we get priego in 14F and prego in 16F). If the Latin word has a long vowel, the outcome was a single vowel (and not a diphthong) in both 14F and 16F (for instance, from Lat. crēdĕre we always obtain credo). Therefore, when the Latin word has a long vowel, the Florentine word is not included in the CrieG group. Of course, there was no way that I could automate a procedure to include a word such as prego in the CrieG group, while leaving aside a word such as credo, because the only difference relies on the Latin etymology. The best I could do was to write two scripts such as: CrieG <- '(?mix)[bcdfgpt] (rie) [^\\b]' and CrieS <- '(?mix)[bcdfgpt] (re) [^0-9] [^0-9]? [^0-9]? \\b', and after, check one by one all the results, discarding the words whose outcome e did not derive from ĕ, ŏ.

Often, I needed to know also the meaning—and not only the etymology—of the word that I was analyzing. The structure of the words ‘stiavo’ and ‘stiano’, for example, is identical, but the former is part of the stiVS group, while the latter is not. Similarly, a word like ‘begli’ is part of the lliS group, while ‘degli’ is not. The four scripts schiVG <- '(?mix)(schi) [aeou] [^\\b]', schiVS <- '(?mix)(sti) [aeou] [^\\b]', lliG <- '(?mix) [aeiou] (lli) \\b' and lliS <- '(?mix) [aeiou] (gli) \\b' account for more results, if compared to the correct ones, and again, a manual check was needed.

When I was finally done counting the correct number of occurrences of the forms that I was searching for, I put the numeric outcomes (= the number of occurrences of every feature) in a separated file, just as I did with the length() function above.

Next, I did chi-squared test on the outcomes, divided by time intervals. The null hypothesis was that the golden and silver forms were randomly distributed over time. I obtained a p-value < 2.2^-16, far below the commonly accepted threshold (0.05).¹⁶ This means that the outcomes were statistically significant, and consequently, that the use of golden and silver forms varies in a non-random way over the years. But the test does not say in what way non-random choices affected Michelangelo’s use of golden and silver forms: that is why I needed to run also the correspondence analysis, “a statistical technique that provides a graphical representation of cross tabulations [...]. Cross tabulations arise whenever it is possible to place events into two or more different sets of categories.”¹⁷

With the function: features_ca <- ca(features) I run the correspondence analysis, and I printed the scree plot (Figure 1).

Given these results, I considered only dimension 1 (time-related), which accounts for 50% of the total.¹⁸ Furthermore, since I was interested in differences among time intervals, I focused on rows (map="rowprincipal", cf. (Michael Greenacre 2007)). At that point, I could finally print the plot of the correspondence analysis (Figure 2).

The data and the subsequent plot show a clear cut-off date around 1530: indeed, values consistently diverge before and after this date. On the left side of Figure 2, together with all the time intervals from 1495 to 1530, are grouped most of the silver forms (= 16F, recognizable by a capital ‘S’ at the end of their names). On the contrary, on the right side of the plot, together with the time intervals from 1530 to 1564, are grouped most of the golden forms (= 14F, recognizable by a capital ‘G’ at the end of the name). Moreover, since the first dimension, corresponding to the horizontal axis, is time-sensitive, I could deduce from the plot a strong separation between silver forms, most of them at the very left side of the plot, and golden forms, most of them—with the only exception of ultimS—at the right side.

However, correspondence analysis could be also sensitive to variation different than time, or gold/silver variation, because the features under scrutiny are not completely independent, but come in pairs of alternative variants (14F vs 16F forms). One could argue that in cases like this, the frequencies of the features are not only determined by the writer’s preference for the one or the other variant, but also by the overall frequency of the lexical items at hand, as is fully explained in (Dirk Speelman 2003). To address this issue, I applied correspondence regression, using the R package “corregp” (cf. (Plevoets 2015)).¹⁹

I then reshaped the data, so to obtain Table 1:

At this point, I performed a correspondence regression of the response variable “feature” in function of the main effect of “time” + the main effect of “measure” + the interaction between “time” and “measure” (cf. (Plevoets 2018): 2–3). The plot in Figure 3 shows that dimension 1 (time-related) and dimension 2 (related to the golden and silvery alternation) account for most of the variation (to be precise, 79% of it).

I then plotted the outcomes of the correspondence regression, as in Figure 4:

The plot confirms the results previously obtained with the correspondence analysis. In the horizontal axis—that accounts for 65% of the total variation—we notice a strong separation between silver (S.) and golden (G.) forms over time. But also the vertical axis shows a connection between two groups of forms:

1. golden forms, 1495–1530 and silver forms, 1530–1560 (top quadrants);

2. silver forms, 1495–1530 and golden forms, 1530–1560 (bottom quadrants).²⁰

The results suggest, one more time, that the key period when Michelangelo started to modify his use of the language is around 1530; moreover, this outcome seems to be independent of the type of lexical items taken into account and their overall frequency.

3 Historical Interpretation of the Statistical Tests

During his life, Michelangelo repeatedly denied his interest in the contemporary debate on language and grammar, and all previous scholars that studied his texts from a linguistic point of view have insisted that his written style is an excellent example of 16F. I hypothesize that—despite his repeated claims to be grammatically ignorant— Michelangelo was aware of the existence of manuals prescribing the Old Florentine and may have used some 14F features more or less deliberately. In this paper, I explored on a quantitative basis Michelangelo’s use of the language in his letters.

Among many parameters, I choose to consider diachrony. First, I split Michelangelo’s letters into documents representing time intervals. Then, I selected the most relevant features that differentiate 14F from 16F and I counted their occurrences in the corpus, aiming to see whether there is a difference in his use of the language, and—if so—whether this difference can be related with historical reasons.

The analysis has shown that most of the silver forms are used before 1530, and most of the golden forms are used after 1530, and that their use varies in a non-random way over time. This result is of extreme interest, because the most important Italian grammar book of that time—the Prose della volgar lingua, written by the renowned humanist Pietro Bembo—was published in 1525. In that book, Bembo prescribes the use of 14F (i.e., the golden forms) as a common language for all Italian people.

So far, there was no evidence that Michelangelo ever read the Prose. Interestingly enough, my data shows that he consciously started using the 14F forms a few years after the publication of that book. I argue that Michelangelo was more informed of the contemporary grammatical dispute than we previously thought. It seems reasonable to affirm that in the years following the publication of the Prose, Michelangelo read a copy of it, and began modifying his written language, following the 14F rules prescribed by Bembo. He did so, because in those days he was no more a simple artisan, as he was at the beginning of his life. On the contrary, in 1525–1530 he already became a public figure, and wanted to emancipate from his humble origins: but, to do so, he needed to polish his language from the most marked 16F phonetic and morphological features, at that time perceived as ‘popular’.

The linguistic evidence showing Michelangelo’s use of the 14F forms prescribed by Bembo not only is a reasonable hypothesis, but also perfectly matches with the historical documentation. Indeed, in the 1520s and 1530s, Michelangelo was in Florence, frequenting the Orti Oricellari together with his friends Donato Giannotti, Battista della Palla and Antonio Brucioli.²¹ They were all devotees to Bembo’s ideals, prone to use the Old Florentine language, and they could easily have introduced Michelangelo to that linguistic system. In particular, in those years, Antonio Brucioli was translating Christian texts into vernacular language: he published the New Testament in 1530, the Psalms in 1531 and the Bible in 1532. We also know that in 1529, when Michelangelo was living in Venice, Michelangelo and Brucioli regularly met.²² Not only then “it is likely that on that occasion, Michelangelo […] has been faced for the first time with the topics of the Protestant Reformation.”²³ Moreover, I would like to emphasize that Michelangelo’s Venetian stay and the reading of Brucioli’s translations may have had some consequences in terms of his linguistic beliefs, too.²⁴

Furthermore, in the following years, Michelangelo and Bembo had both stayed at the papal court in Rome. Unfortunately, in the absence of documents witnessing Bembo and Michelangelo friendship, we cannot say much about it, but we are supported by Vasari’s words. In the Life of Michelangelo, he states:

The illustrious Cardinal Polo was his close friend, and Michelangelo loved his virtue and goodness. Other friends were Cardinal Farnese and Santa Croce, who later became Pope Marcellus II; Cardinal Ridolfi and Cardinal Maffeo and Sir Bembo, Carpi and many other cardinals and bishops and prelates that we do not mention.²⁵

Despite there is no evidence of it, the two of them are likely to have discussed grammar and language, and it is possible that Michelangelo showed some kind of interest in the recently printed Prose della volgar lingua, the grammar book that was revolutionizing the entire linguistic debate in the whole Peninsula.

Indeed, the preference given by the sculptor in those very years for the use of linguistic features characteristic of the 14F system may be a reflection of his learned dissertations and his increasing social status, and consequently, might reveal his wish to align his language with the 14F grammatical rules prescribed by Bembo in 1525. Therefore, the historical documents that witness his frequentations—starting from 1520—with the key players of the sixteenth-century grammatical dispute, confirm and support the results of the correspondence analysis and the correspondence regression.

Likewise, a few years later (1542) Michelangelo asked his friends Donato Giannotti and Luigi del Riccio to amend the language of his poems:

Sir Luigi, you who have the spirit of poetry, I beg you to shorten and improve one of these madrigals, which at the moment is imperfect, because I must give it to a friend of ours.²⁶

As showed in (Valenti 2019), he was probably asking for a review of the linguistic features that did not match with Bembo’s grammatical rules. This is the last piece of evidence that call into question the old assumption that Michelangelo was unaware of the grammar books prescribing the use of fourteenth-century Florentine language. In fact, his linguistic choices did not always reflect the contemporary use and, sometimes, he was more inclined to employ the archaic forms than we would expect.

Michelangelo Buonarroti once defined himself “grammatically mistaken.”²⁷ Maybe he did not know all the rules of 14F listed in the Prose, as other people of his time did, but this analysis shows that he was far from being completely unaware of them.

Acknowledgements

I am deeply grateful to Pietro Mercuri (Sapienza University, Rome) for his help with the statistical tests. The codes that I used for my analyses are available here: https://github.com/gianlucavalenti/A-Corpus-Based-Approach-to-Michelangelo-s-Epistolary-Language.