Daejeon, July 27–31
The city as a literary topos “is certainly not an invention of modernity” (Braungart 2008, 16, transl. ours), but in the 19th century, full-blown industrialisation and urbanisation (Fisher 1988, 107) made the modern metropolis and the experience of urban spaces a rather dominant theme. In the German-language context, literary imaginations of the city, addressing dimensions of space, sociality, and affect, concentrate especially on ‘Vienna’ and ‘Berlin’.
A Computational Literary Studies perspective on urban space can build on several studies that address the relationship between emotions and aspects of space, especially concreteness. Kröncke et al. (2022, 2023, 2024) examine text-encoded emotions and concreteness in poetry from Realism to Modernism, Grisot and Herrmann (2023, 2025) examine emotional encoding of types of space in German-language fiction around 1900, while a host of studies observes a historical increase in linguistic concreteness in English-language prose (Heuser 2019; Snefjella, Généreux, and Kuperman 2019; Sun and Wang 2022).
This article examines text-encoded emotions and concreteness in spatial descriptions of ‘Vienna’ and ‘Berlin’ in German-language narrative prose fiction. Asking about diachronic change, it situates both big cities within their distinct developments of literary modernism around 1900. Building on prior literary-historical work, our initial hypothesis is that within the time frame, spatial descriptions of the big city become overall more concrete. We also assume that positive emotions encoded in the texts decrease.Arno Holz, a proponent of naturalism, pleaded for a more concrete and creative description of reality in literature (cf. Kanz 2019, 348). In addition, we ask about two specific ‘modernisms’ that emerged around 1900, whereby we link the degree of concreteness and emotional encoding to spatial descriptions of the two respective cities. Finally, we introduce the “Big City Prose Corpus” (GroProKo) that includes literary prose texts that depict or mention either Berlin or Vienna. Emotions and lexical concreteness are identified using automatic annotation.
Literary texts vary in their degree of concreteness and also their emotional profiles diachronically and synchronically, which poses interesting questions about the representation of space in terms of diegesis as well as style. Sentiment analysis, the computational analysis of text-encoded emotions, has become a standard instrument in CLS (Haider et al. 2020; Dennerlein, Schmidt and Wolff 2023; Grisot and Herrmann 2023; Herrmann and Lüdtke 2023; Kröncke et al. 2024). Snefjella, Généreux and Kuperman (2019) and Sun and Wang (2022) show a diachronic increase in the concreteness of English from the nineteenth to the twentieth century. Herrmann (2018) examines the relationship between metaphor, word classes, and vividness, showing that a type of concreteness is present in vivid metaphors involving content words, but also in texts with low metaphor density that appear to belong to a different genre.
In terms of method, lexicon-based detection of sentiment, as well as concreteness, has to deal with word-level ambiguity because the dictionaries are not context-sensitive (Löhr 2022). However, both sentiment and concreteness analysis have been evaluated to produce good-enough results. For sentiment lexicons, coverage is a typical measure to evaluate validity (cf. Grisot and Herrmann 2023). The validity of concreteness lexicons has been evaluated by correlating human concreteness ratings with embedding-based scores (Bizzoni, Feldkamp, and Nielbo 2025).
Urban fiction writing around 1900 explores the social, political, and cognitive effects of industrialisation and urbanisation, where cities function as spaces of action and imagination (Williams 1973; Schutte and Sprengel 1987; Fisher 1988; Berbig et al. 2011; Berbig and Göttsche 2013). In German-language literature, Berlin and Vienna in particular are each associated with their own cultural and poetic ‘modernism’ (Wunberg 1993).
Between 1871 and 1912, Berlin experienced massive growth (Reulecke 1985; Berbig et al. 2011, 2), marked by the increasingly pressing ‘social question’ (Bodenschatz 2011). As an ‘industrial hub’, Berlin did not appear very aesthetically pleasing (Schutte and Sprengel 1987; Berbig et al. 2011). By the late nineteenth century, poorer milieus became a key thematic focus of especially naturalist texts, characterised by a high degree of detail and concreteness. The ‘Berliner Moderne’ demanded a departure from Poetic Realism and its idealised representation of reality (Kanz 2019).
From roughly 1850 to 1918, Vienna was shaped by the economic and social tensions of the disintegrating Habsburg monarchy (Huyssen 1998; Irsigler and Orth 2015). Urbanisation and industrialisation fueled population growth and social tensions (Oei 2013; Lorenz 2007). Around 1900, Vienna’s heterogeneous cultural and literary landscape (Lorenz 2007, 4; Irsigler and Orth 2015, 9) became associated with melancholic topoi (e.g., Oei 2013; Wolf 2015), incarnated by the fin de siècle and the “décadence”, an atmosphere of decline (Irsigler and Orth 2015, 12). These literary “moods” are specifically rendered in descriptions of space (Niefanger 2011, 52).
As this Viennese literature programmatically distances itself from Naturalism and Realism (Lorenz 2007, 31; Irsigler and Orth 2015, 27) with a more symbolic poetics, we assume that spatial descriptions in Viennese literature around the turn of the century are more abstract and more negative than those in Berlin literature.
We introduce GroProKo, which consists of fictional German-language prose texts published between 1775 and 1935 (Figure 1) and bears reference to Vienna and/or Berlin. It comprises N=252 full literary texts (N=26.11 million tokens) by a total of N=109 authors (n=89 male, n=20 female).
For the corpus, we selected prose texts classified as urban writing in literary research, as well as texts with particularly frequent mentions of “Berlin” or “Vienna” from a larger corpus with N=2,635 texts (cf. Kababgi and Herrmann 2023). Historical spellings were normalised using CAB (Jurish 2012). Of the N=252 texts, n=131 deal with Berlin and n=121 with Vienna.
Spatial descriptions were automatically extracted from the corpus using Qwen2.5-32B-Instruct (Yang et al. 2024) as described and recommended by Sitter et al. (2025). All sentences describe static spatial conditions without any immediate actions.
Sentence-level concreteness scores were calculated for the spatial descriptions using the concreteness lexicon of Köper and Schulte im Walde (2016). We computed the mean concreteness score of all lemmatised words per sentence (excluding stop-words) to account for varying sentence length and stop-word usage. Text-encoded emotions and valence were calculated using SentiArt (Jacobs 2019), which was specifically developed for literary texts (Jacobs 2017; Jacobs and Kinder 2019; Jacobs et al. 2020). The sum of SentiArt values per sentence auto-normalises since SentiArt scores can be both positive and negative. The length of the sentence does not matter, since a large number of words indicating a certain trend show up this way. A low value for either valence or any emotion signifies low textual expression or absence of the given variable, while a high value describes strong textual expression of the text-encoded emotion or valence.
For the statistical analysis, we fitted a generalized regression model (Friedman et al. 2010) using R (R Core Team 2025). The model included city and the publication year as fixed factors with an interaction between these predictors.Statistical analysis used R-4.5.2 with the glm function of the stats-modul (v.4.5.2) with the following specification “x ~ pubyear * city”. The interaction term was investigated using Type III ANOVA in the ‘car’ package (Fox and Weisberg 2019).
From the N=252 texts, we extracted N=7,985 spatial descriptions, of which n=5,112 sentences originate from Berlin texts and n=2,873 sentences from Vienna texts (see Figure 2).
For the dependent variables concreteness, valence, joy, anger, disgust, and surprise, we observe significant diachronic differences. For sadness and fear, no significant diachronic differences can be observed. For the variables concreteness, valence, disgust, and fear, there are significant differences between the cities and a significant interaction between publication year and city (Table 1).
Table 1: Results of the statistical analysis. CI = confidence interval. The column “Variable” lists the variable and the associated factors that differ statistically. “City” indicates statistical differences between Vienna and Berlin, “pubyear” shows diachronic differences, and “city × pubyear” shows differences in diachronic change between the cities. Significant differences are highlighted: p-values below α=0.01 **, p-values below α=0.05 *.
| Variable | Mean Berlin | Mean Vienna | CI 95% | p |
| Concreteness … | 5.16 | 5.10 | ||
| … city | 0.26 – 3.66 | 0.024 * | ||
| … pubyear | 0.00 – 0.00 | <0.001 ** | ||
| … city × pubyear | -0.00 – -0.00 | 0.020 * | ||
| Valence … | 4.65 | 5.49 | ||
| … city | 22.45 – 52.30 | <0.001 ** | ||
| … pubyear | -0.01 – -0.00 | 0.024 * | ||
| … city × pubyear | -0.03 – -0.01 | <0.001 ** | ||
| Sadness … | 2.37 | 2.47 | ||
| … city | -14.07 – 15.85 | 0.907 | ||
| … pubyear | -0.00 – 0.00 | 0.914 | ||
| … city × pubyear | -0.01 – 0.01 | 0.918 | ||
| Joy | 3.44 | 3.7 | ||
| … city | -17.57 – 13.42 | 0.793 | ||
| … pubyear | -0.01 – -0.00 | 0.034 * | ||
| … city × pubyear | -0.01 – 0.01 | 0.764 | ||
| Anger … | 2.59 | 2.84 | ||
| … city | -6.47 – 4.41 | 0.710 | ||
| … pubyear | -0.01 – -0.00 | <0.001 ** | ||
| … city × pubyear | 0.00 – 0.00 | 0.651 | ||
| Disgust … | -0.54 | -0.72 | ||
| … city | -6.47 – 4.41 | 0.005 ** | ||
| … pubyear | -0.01 – -0.00 | 0.352 | ||
| … city × pubyear | -0.00 – 0.00 | 0.005 ** | ||
| Surprise … | 0.61 | 0.53 | ||
| … city | -6.47 – 4.41 | 0.112 | ||
| … pubyear | -0.01 – -0.00 | 0.014 * | ||
| … city × pubyear | -0.00 – 0.00 | 0.109 | ||
| Fear … | 1.71 | 1.70 | ||
| … city | -35.11 – -8.72 | 0.001 ** | ||
| … pubyear | -0.00 – 0.00 | 0.749 | ||
| … city × pubyear | 0.00 – 0.02 | 0.001 ** |
The regression lines for both cities decrease in parallel for joy and anger, while sadness remains constant across the investigated timeframe. The regression lines for concreteness, valence, disgust, and fear intersect from around 1850 onwards and differ significantly. Overall, values for concreteness and disgust increase, but valence decreases. Fear increases for Vienna and decreases for Berlin (Figure 3). Diachronic change in valence, disgust, and fear is more pronounced in the Vienna subset. The Berlin subset shows a stronger increase only for concreteness, with an intersection already before 1850. The lines for valence and disgust are the only ones to intersect again after 1900.
In line with our first hypothesis, for spatial descriptions, we observe a general shift in concreteness (increase) and negative emotions (increase), but with slightly different patterns for the two cities: The increase in negative emotions is more drastic for Vienna (see fig. 2 B, p=<0.001). The increase in spatial concreteness is stronger for Berlin (see fig. 2 A, p=0.020). In line with our second prediction, Viennese descriptions are overall less concrete than their Berlin counterparts. By contrast to our third expectation (melancholy), we do not observe any significant change in sadness (p=0.918). However, the decrease in valence (p=<0.001) and increase in fear (p=0.001) in both cities implies that melancholy does not manifest solely as sadness.
For disgust (p=0.005) and fear (p=0.001), there is a significant interaction between publication year and city, with Vienna’s gradient being steeper. This increase in disgust and fear may be cautiously associated with the declining Habsburg monarchy as a literary theme.
Valence decreases for both cities, again with a more pronounced decline for Vienna (p=<0.001). The results overall suggest that texts depicting Vienna show a stronger decline in positive emotions between 1775 and 1935. A finding that may be associated with décadence and a rejection of Naturalism.
Further studies on emotions and concreteness are needed to model genre and the distinction between literary prestige (Herrmann 2018; Kröncke et al. 2023), as well as literariness and fictionality as predictors. Follow-up studies could also take into account the texts' canonicity and underlying poetics, as well as aspects of ‘nationality’ as a historical factor within the German-speaking countries. Analysis procedures optimised for semantic ambiguity should be used. Alongside GroProKo, we are currently developing a corpus of ‘anti-modern’ texts from the same period that explicitly oppose the big cities (cf. Rossbacher 1975). A complementary study devoted to spatial descriptions of the village is in preparation.