Measuring the style of chick lit and literature

1. Introduction
Novelistic genres have certain formal conventions. By close reading novels of various genres, structural features such as the differences in theme, motifs and plot can be observed. According to Jockers (2013), there seem to be stylistic differences, caused by the author’s linguistic choices, between genres as well.

In this paper I examine the stylistic fingerprints of chick lit and literature in terms of the most frequent words. The focus on the styles of these particular genres relates to the overarching question of my research: does the perception of literariness (e.g. the dichotomy between “highbrow” and “lowbrow” genres) correlate with certain linguistic characteristics, or with the degree of variety of these in the style? The aim of this pilot study is to explore how well successful stylometric methods could be applied as a starting point for such a comprehensive question.

2. Background
Stylometrists who study linguistic patterns in fiction typically focus on classification tasks, e.g. authorship attribution or text genre detection. The latter studies usually examine how well certain texts can be identified into pre-defined classes; for instance editorials, newspapers and literature (Stamatatos et al., 2000). Stylometric studies of novelistic genres (e.g. Louwerse et al., 2008; Ashok et al., 2013) seem to be scarce. The most extensive study is performed by Matthew Jockers (2013) and his colleagues at the Stanford Literary Lab (Allison et al., 2011). They examine to what extent formal conventions can be detected at the level of the high-frequency function words. Jockers (2013) concludes that genres to a certain degree have measurable linguistic fingerprints, and that linguistic decisions of the authors seem to be dependent upon their genre choices. An interesting next step would be to analyze and interpret these fingerprints, as has been done for authorial markers by Burrows and Craig (2012). In this paper I adopt their approach to examine the stylistic differences between chick lit and literature.

A previous study into the deep syntactic structures of the two genres shows that literary authors tend to use more complex sentences and employ a more descriptive language, whereas chick lit to a greater extent resembles colloquial speech (Jautze et al., 2013). In the current study I complement this syntactic characterization with a stylometric analysis of the function words. These words relate to syntactic structure because they add grammatical information by organizing and connecting the content words. The question arises if the most frequent words (MFWs) differentiate between the two genres. And if so, do these linguistic patterns give more insight into the two genre styles?

3. Materials and method
According to Jockers (2013) it is hard to distinguish linguistic fingerprints that are related to the time of writing from actual genre signals. This means that when one wants to examine genre fingerprints, the chronology factor must be ruled out as far as possible. My corpus therefore comprises 32 original Dutch novels (16 literary and 16 chick-lit) of the last two decades.

In order to computationally examine the style of the two genres I start with the stylometric approach to search for the style markers. The stylo package in R compiles a word frequency list for the entire corpus (Eder et al., 2013). Then, I want to explore the language patterns to characterize the two genre styles. Egbert (2012) argues that lexicogrammatical features can be captured in three dimensions of discourse presentation. Two of these dimensions I will adopt in this analysis in order to analyze the linguistic patterns: (i) description versus thought representation and (ii) dialogue versus narrative.

4. Results
Figure 1 shows that the titles per author as well as the two genres cluster together (the abbreviations indicate the pre- defined genres). According to Jockers (2013), men and women tend to use different (function) words, so one might have expected the female literary authors in my corpus to cluster together with the female chick-lit writers. My results indicate the opposite, which suggest that (in this corpus) genre signals precede gender signals. It is striking however that the chick-lit writers are more grouped together than the literary authors. This indicates that there is more variation within the literary writing styles.

Fig. 1: A Bootstrap Consensus Tree showing average similarity of texts based on the frequencies of 100-1000 MFW.

In order to examine linguistic patterns behind this genre-distinction, the word frequencies are analyzed. A Principal Components Analysis uses the MFWs as variables according to which the texts are correlated in a matrix. Figure 2 shows that the 100 MFWs map the genres in separate areas of the graph, except for chick-lit writer Wilma Hollander. She sides with the literary authors.

Fig. 2: A PCA showing the plotting of texts based on the weightings of 100 MFW.

The two components of the PCA together account for 43.7% of the variation between the novels. The word-variables have their own weightings for each component according to which the texts are scored in the matrix (e.g. Figure 3). In the previous study by Jautze et al. (2013) the novels were parsed with the Alpino parser (Bouma et al., 2001). The parts-of- speech tags made it possible to separate homographic word forms, as in zijn (‘his’) and zijn (‘are’).

Fig. 3: A PCA showing the plotting of 100 word-variable weightings.

POS tags Translation tags
Vnw Pronoun
Ww Verb
Bw Adverb
Vg Conjunction
Vz Preposition
Lid Determiner
Adj Adjective
N Noun
Tsw Interjection
With regard to Egbert’s dimensions, it can be argued that the literary authors employ more descriptions and narratives, whereas in chick lit more thought representations and dialogues are used. Indicative for the descriptive dimension is the high amount of prepositions, the use of determiners and the demonstrative die (‘that’). Prepositions express spatial or temporal relations between subjects and/or objects, and therefore are used for detailed-oriented description. Along with the use of determiners and demonstratives, this indicates that the literary authors use relatively more nouns. These findings can be underlined by the results of Jautze et al. (2013), that show that noun phrases and prepositional phrases occur more frequently in the literary books than in the chick-lit novels of our corpus.

Other frequent “literary” function words in the PCA are third person pronouns such as hij and hun (‘he’ and ‘their/them’), indefinite pronouns such as iets and alles (‘something’ and ‘everything’) and verbs in the past tense. According to Egbert (2012), these linguistic features belong to the narrative dimension. Especially the past tense verbs indicate that the literary narrators describe events. The chick- lit writers on the other hand, employ more present tense verbs, and first and second person pronouns such as ik, mij and jij (‘I’, ‘me’ and ‘you’). These, as well as the demonstratives dat and daar (‘that’ and ‘there’), are argued to be indicative for the dialogue dimension.

Moreover, at the chick-lit side of the plot a lot of words are mapped that relate to the dimension of thought representation. Function words like the mental verb weet (‘know’), the indefinite pronoun veel (‘many’), the affective adjectives heel and goed (‘very’ and ‘good’), the possibility modal kan (‘can’) and the likelihood adverb misschien (‘maybe’) offer insight into the narrator’s or character’s psyche. The chick-lit authors also employ certain adverbs (maar, toch) that can cause an emphatic effect. It could be compared with ‘there are only seven’. It shows a character’s or narrator’s involvement, and it belongs to a more colloquial language register.

5. Conclusion
The results of this paper show that stylometric analysis can be used in stylistic research of literary genres. The linguistic patterns detected in this small corpus suggest that the literary authors have a more detail-orientated descriptive style when compared to the chick-lit style, which tends to be more informal and involved. The preliminary results offer a variety of clues to further research. In the next stage I would like to explore for instance word n-grams and parts of speech in a corpus that is expanded with several other “highbrow” and “lowbrow” genres.

Acknowledgements
I am grateful to my supervisor professor Karina van Dalen- Oskam and to Corina Koolen and Andreas van Cranenburgh for reading my drafts, and to Andreas for assisting with tagging the parts of speech in the novels.

References
Chick-lit novels humorously address the challenges of young urban female protagonists.

his study is part of The Riddle of Literary Quality Project. In this project we explore the assumption that formal characteristics play a role in the aesthetic appreciation of novels. Cf. literaryquality.huygens.knaw.nl

This is the same corpus as has been studied in Jautze et al. (2013). Ideally, female and male writers should be equally represented. But since the chick-lit novels were all written by women, this was not possible.

This Bootstrap Consensus Tree is a mean of ten cluster analyses, varying from 100-1000 MFWs with an increment of 100. The corpus is culled at 100%, which means that words that are unique for individual texts are removed.

Allison, S., Heuser, R., Jockers, M., Moretti, F. and Witmore, M. (2011). Quantitative Formalism: An Experiment. In Pamphlets of the Literary Lab 1,litlab.stanford.edu/LiteraryLabPamphlet1.pdf, (accessed on 24 October 2013).

Ashok, V.G. , Feng, S., and Choi, Y. (2013). Success with Style: Using Writing Style to Predict the Success of Novels. In Empirical Methods on Natural Language Processing, Seattle. 1753-1764,aclweb.org/anthology/D/D13/D13-1181.pdf, (accessed on 28 October 2013).

Bouma, G., Van Noord, G. and Malouf, R. (2001). Alpino: Wide-coverage computational analysis of Dutch. In Language and Computers, 37 (1). 45–59.

Burrows, J. and Craig, H. (2012). Authors and Characters. In English Studies 93 (3). 292-309.

Eder, M., Kestemont, M. and Rybicki, J. (2013). Stylometry with R: a suite of tools. In Digital Humanities 2013: Conference Abstracts. Lincoln (NE). 487-489.

Egbert, J. (2012). Style in nineteenth century fiction. A Multi- Dimensional analysis. In Scientific Study of Literature 2 (2). 167-198.

Jautze, K., Koolen, C., Van Cranenburgh, A. and De Jong, H. (2013). From high heels to weed attics: a syntactic investigation of chick lit and literature. In Proceedings of the Workshop on Computational Linguistics for Literature. Atlanta (GA). 72-81,aclweb.org/anthology//W/W13/W13-1410.pdf.

Jockers, M. L. (2013). Macroanalysis: Digital Methods and Literary History. Illinois: University of Illinois Press.

Louwerse, M., Benesh, M.N. and Zhang, B. (2008), Computationally discriminating literary from non-literary texts. In: Zyngier, S., Bortolussi, M., Chesnokova, A. & Auracher, J. (Eds.), Directions in Empirical Literary Studies. Linguistic Approaches to Literature 5, Amsterdam. 175-191.

Stamatatos, E., Fakotakis, N., & Kokkinakis, G. (2000). Text Genre Detection Using Common Word Frequencies. In Proceedings of the 18th conference on Computational linguistics (2). Stroudsburg (PA). 808–814.