Claudine at the workshop: the impact of Willy and his secretaries on Colette's writing

What was the influence of her first husband Willy and his literary workshop on the French writer Colette? Her opinion varied greatly, depending on when she spoke. Asked just after the publication of the Claudines, a series of books published from 1900 to 1903 initially under her husband Willy's name alone, Colette paid lip service to her contribution: "[Willy] so obstinately cries out for my participation [in the Claudines] that I have to bow down". But in 1936 in her autobiographical work Mes Apprentissages, and until her death in 1954, Colette changed her tone and insisted on Willy's minor role in the writing of these books, until her first husband's name disappeared from the cover of the novels. When Colette was asked if Willy had helped her to write these texts, she replied in 1949: "Rather by indications, but that cannot be called help... ". Which Colette should we believe? 

In this paper, we analyse texts that we have ocerised and corrected. Using artificial intelligence profiling techniques, we show that the novel Claudine à l'école bears the influence of Willy and his workshop.

Claudine à l’école: a collaborative work?

Sidonie Gabrielle Colette, known as Colette (1873-1954), took her first steps on the French literary scene in the shadow of her first husband Henry-Gauthier Villars (1859-1931), better known under his pseudonym Willy. Between 1900 and 1903, four novels belonging to the Claudines series were published and signed by Willy: Claudine à l'école (1900), Claudine à Paris (1901), Claudine en ménage (1903), Claudine s'en va (1903). Willy never concealed the fact that Colette had participated in the writing of these books, as Colette acknowledged in Mes Apprentissages. The use of his name is explained both by commercial reasons (Willy was a successful author) and by social reasons: as a woman in 1900, Colette might have encountered many difficulties in publishing a novel under her own name. Willy's methods of literary production were well known to his Parisian colleagues and friends. In 1900, it is well known that Willy employed numerous "secretaries" or "collaborators" to write "his" books. Colette describes their flat as a 'literary workshop', where several pens wrote novels, meticulously revised, rewritten and commented on by Willy. In the course of his literary career, it is estimated that about fifty collaborators were employed by Willy. The draft manuscript of Claudine à l'école would also have been 'processed' by this workshop; and after being revised by Willy, it would have been reduced by half (Caradec, 2004).

Nevertheless, we do not know exactly which part of the text was written by Colette and which part was rewritten (or even written) by Willy or his collaborators. To what extent did Willy influence the first literary texts written by Colette? What role did he actually play in the writing of these novels?

A methodological challenge

Analysing the relative contributions of Colette, Willy and his secretaries poses numerous methodological problems. First of all, texts that can be attributed with certainty to Willy alone are very rare - even more so if we restrict ourselves to the novelistic genre. Finding texts written by individual participants in Willy's workshop is also a complex task. These writers are very numerous, not all of them have pursued a literary career, and some of them have probably continued to write in collaboration with other authors. It is therefore very difficult to treat the Claudines as other cases of collaborative authority (Plechac, 2019; Cafiero & Camps, 2021): training an artificial intelligence to recognise the style of each of the candidates, and then applying a rolling stylometry algorithm (Eder, 2016) is a challenge in the short term, due to the lack of reliable texts available.

Moreover, if Colette is telling the truth, Willy's main influence on these texts would come from comments and suggestions made on a text written and then corrected by herself. If this is indeed the case, a classical stylometric approach would fail: it would only detect Colette's hand, but not her influences. (Cafiero & Camps, 2019).

Profiling

If it is not easy to identify Willy's writing in Colette's text, we can try to identify his influence in other ways. The reasoning is simple: compare texts written by authors in Willy's workshop with those they wrote on their own, outside the workshop, as close as possible to their publication date. What is the difference between these texts? To a first approximation, this is what we will call the Willy effect. It includes Willy's corrections, advice or writing passages, but is not limited to them. Others may well have helped their colleagues to bring their manuscript closer to Willy's expectations - by editing, giving advice etc. 

To measure this effect, we use a profiling logic (Bevendorf et al., 2021; Daelemans et al. 2019; Mikros, 2017). We train an artificial intelligence (Support Vector Machine (SVM)) to distinguish between texts written by certain authors in collaboration with Willy and texts written by the same authors outside Willy's workshop. The training is done on a corpus of novels by Paul Acker, Maurice Edmond Sailland (known as Curnonsky), Jeanne Marais, Jean de Tinan, Léo Trézenik and Pierre Veber. The SVM takes as input a combination of the most frequent character bi-grams, character tri-grams, words, and word bi-grams. The optimal parameters are deduced by cross-validation. A polynomial kernel is adopted, which performs better here than various linear kernels. Low variance features are removed to reduce the computation time. The performance of this profiling method is satisfactory (F-score: 0.89). 

The texts to be evaluated (Colette's novels) are then divided into sections of a thousand words. For each section, the SVM is asked whether the text seems to be influenced by a Willy effect or not.

References

• BEVENDORFF, Janek, CHULVI, Berta, PEÑA SARRACÉN, Gretel Liz De La, et al. Overview of PAN 2021: Authorship Verification, Profiling Hate Speech Spreaders on Twitter, and Style Change Detection. In : International Conference of the Cross-Language Evaluation Forum for European Languages. Springer, Cham, 2021. p. 419-431

• CAFIERO, Florian et CAMPS, Jean-Baptiste. Why Molière most likely did write his plays. Science advances, 2019, vol. 5, no 11, p. eaax5489.

• CAFIERO, Florian et CAMPS, Jean-Baptiste. ‘Psyché’as a Rosetta Stone? Assessing Collaborative Authorship in the French 17th Century Theatre. Proceedings http://ceur-ws. org ISSN, 2021, vol. 1613, p. 0073

• CARADEC, François. Willy: le père des Claudine. Fayard, 2004.

• DAELEMANS, Walter, KESTEMONT, Mike, MANJAVACAS, Enrique, et al. Overview of PAN 2019: bots and gender profiling, celebrity profiling, cross-domain authorship attribution and style change detection. In : International conference of the cross-language evaluation forum for european languages. Springer, Cham, 2019. p. 402-416.

• EDER, Maciej. Rolling stylometry. Digital Scholarship in the Humanities, 2016, vol. 31, no 3, p. 457-469.

• MIKROS, George K. Blended Authorship Attribution: Unmasking Elena Ferrante Combining Different Author Profiling Methods. UPPADO, 2017, p. 85.

• PLECHÁČ, Petr. Relative contributions of Shakespeare and Fletcher in Henry VIII: An analysis based on most frequent words and most frequent rhythmic patterns. Digital Scholarship in the Humanities, 2021, vol. 36, no 2, p. 430-438