Introduction

In this paper,

we explore how to uncover stylistic forms of literary influence using both stylometric analysis and research into a broader range of contextual information than is usual in existing computational analyses of style. We argue that there is a danger of focusing too much, or even solely, on stylometric techniques. For instance, the analysis of influence between Percy Bysshe and Mary Shelley by Suddaby and Ross (2023) uses stylometry only and ignores valuable criticism offered by philologists and literary scholars, including Robinson’s The original Frankenstein (2008) edition. Instead, we propose a new model for tracing literary influence by synthesizing genetic criticism, literary studies, cognitive stylistics, computational literary studies, and psychology.

The analysis of literary influence has a long history (e.g. Bloom, 1973), yet has received new attention afforded by the arrival of new computational techniques. Stylometric analysis has been used fruitfully by many scholars on the work of many writers, past and present, to analyse different kinds of collaboration in writing and translating (e.g. Eder / Rybicki 2012; Rybicki / Heydel 2013; Rybicki 2014; Van Dalen-Oskam 2014; Eder 2016). Whereas in collaboration contributions are more concrete and measurable, influence is more diffuse and more difficult to prove. Here, we hope to take steps in the direction of a multi-disciplinary approach that helps understand the nature of literary influence.

Case study and methods

Our case study is the influence between Margaret Atwood and Naomi Alderman, which could be the effect of the Rolex Mentor and Protégé collaboration (Coady 2013) they partnered up during 2012–13. The case study was prompted by the results of anexploratory stylometric analysis of a random set of contemporary English-language novels, in which we found that Margaret Atwood’s Shakespeare-retelling Hag-Seed (2016) and Naomi Alderman’s novel The Power (2016) are stylistically very similar to each other compared to the other novels in our ad hoc corpus. This led us to investigate the background and history of their connection, the results of which we then tried to verify using a set of stylometric approaches. We used the Stylo package for R for cluster analysis, principal components analysis, rolling classify, bootstrap consensus trees (Eder et al. 2016), and Gephi to visualize these results in a network analysis (Bastian et al. 2009). Our main corpus included all five novels by Alderman, eight by Atwood, one known co-authored story by Atwood and Alderman entitled The Happy Sunrise Zombie Home (2013), and one collaboration of Atwood and a host of other authors, Fourteen Days (2024).

Stylometric results

Both Atwood and Alderman are well-known for their dystopian and speculative fiction. To rule out that genre played a role in our serendipitous finding, the first step was to check whether Atwood’s Hag-Seed and Alderman’s The Power also end up being very similar in a corpus consisting of our main corpus of Atwood and Alderman works added with a selection of dystopian novels from 1976-2024. This proves to be the case.

Figure 1: Dystopian corpus, cluster analysis 100 MFW, settings: English (ALL), Cosine Delta, no culling.

Figure 1 shows the results of a cluster analysis of the 100 most frequent words, in which Hag-Seed and The Power end up in the same cluster, with three of Alderman’s novels very far away from Atwood’s work. Using 1000 MFW, Atwood’s works make up one of the two main clusters, and Alderman’s novels can be found in the cluster with works by the other authors and the two collaborative texts (graph not shown).

When we apply principal components analysis (PCA), with 100 MFW Hag-Seed and The Power again are closest to each other (Fig. 2). With 1000 MFW, The Power is closest to Atwood’s The Handmaid’s Tale (graph not shown). This suggests a consistent closeness of The Power with more than one work by Atwood.

Figure 2. Dystopian corpus, PCA, correlation version, settings: English (ALL), no culling.

Next, we took a closer look at Alderman’s and Atwood’s collaborative project started early on during their Rolex Mentor and Protégé collaboration, The Happy Sunrise Zombie Home (2013), to verify whether we can easily trace the authors’ individual style. The story was inspired by the success of a running app authored by Alderman, the highly successful game Zombies, Run! (2012). For three months, the writers emailed each other alternating chapters without knowing what would come next and published them on theinteractive Byliner and Wattpad platforms. To investigate potential literary influence, we used the Stylo function Rolling classify. We added milestones to our test set, indicating the start of each of the fifteen chapters (a – n). We selected Cosine Delta and svm forour measurements, using a slice window of 5000 with an overlap of 4500. The result did show some alternations, but not as many as we expected for the different chapters. We then decided to try out a slice window agreeing with the average chapter length, 1,765 tokens. The overlap was 1,665 tokens. Figure 3, showing the results for the 100 MFW, looks a bit messy. Figure 4, for the 1000 MFW looks surprisingly clear. Zooming in on the collaboration between the two authors showed that Atwood and Alderman at the time of writing about zombies retained a measurable difference in their writing styles.

Figure 3. Rolling classify, 100 MFW, Cosine Delta, svm, slice window 1765, overlap 1665.

Figure 4. Rolling classify, 1000 MFW, Cosine Delta, svm, slice window 1776, overlap 1665.

Zooming out again, we created a network analysis of the dystopian corpus we used for the first step, using Gephi and the ForceAtlas2 layout. We made a bootstrap consensus analysis for two ranges: 10-100 MFW with an increment of 10 (Fig. 5), and 100-1000 MFW with an increment of 100 (Fig. 6). The different colours indicate the different clusters with modularity resolution 1.0. As we can see, Hag-Seed and The Power are in the same cluster in both Figure 5 and Figure 6.  This could perhaps be the result of the intense collaboration by Atwood and Alderman in 2012-13.

Figure 5. Network visualization BCT Cosine Delta 10-100 MFW, incr. 10.

Figure 6. Network visualization BCT Cosine Delta 100-1000 MFW, incr. 100.

Additional observations and discussion

Based on our measurements, we hypothesize that Alderman’s The Power (2016) is heavily shaped by the subject of Atwood’s dystopian classic The Handmaid’s Tale (1984) while we also see some Atwoodesque influence at the level of Alderman’s style. Andalthough Atwood’s Shakespeare retelling Hag-Seed (2016) does not bear much influence from Alderman’s style, Atwood’s The Heart Will Go Last (2015) is influenced by Alderman’s innovative play with new technologies. Alderman’s The Future (2023), published a decade after her mentorship by Atwood, shows that Atwood’s influence wanes. This partnership challenges established theories of collaboration, including Pennebaker’s synergy and average-person hypotheses (Pennebaker 2011), and Eder’s palimpsestic model (Eder 2016). Rather than merging into a newly unique, blended or averaged style, Atwood and Alderman’s work reflects a sustained, reciprocal aesthetic exchange.

To cross-check this extraordinary partnership, we also analysed another Rolex collaboration between Toni Morrison and Julia Leigh, where we saw that collaboration does not always have a strong effect on the writing style of authors. And looking at the other title in which Atwood collaborated with others, a Decameronian project that collates 36 different writers into one narrative, Fourteen Days (2024), we see that it is not co-editor Atwood’s stylistic influence that shapes this novel, but that Atwood’s co-editor Preston has a major, shaping hand in the resulting fiction.

Conclusion

In short, our findings reveal a profound mutual influence on personal, thematic, formal, and stylistic levels between Atwood and Alderman during their output following their Rolex collaboration, extending beyond the mentorship period. We can only know this for certain when we invoke contextual information: our comprehensive, multidisciplinary method offers a more nuanced, persuasive and holistic account of literary influence

In future work, to test and finetune our new approach for the analysis of literary influence, we will proceed to examine another writer with ostensibly Atwoodesque traits, Christina Dalcher, whose work in fact does not evidence a stylistic convergence, or even imitation. We expect this approach will have broader implications for understanding literary influence. Our work shows that authors do not always experience anxiety about being influenced: an open mind that welcomes collaboration can just as much lead to creative brilliance.