The rise and fall of biodiversity in literature: A comprehensive quantification of historical changes in the use of vernacular labels for biological taxa in Western creative literature

A free Plain Language Summary can be found within the Supporting Information of this article. Our planet is losing biodiversity at unprecedented rates due to land-use change, direct exploitation, climate pollution and invasion exotic species (Cardinale et al., 2012; IPBES, 2019; Millennium Ecosystem Assessment, 2005; Tilman, 1999). In Western countries, for example, loss had started already in second half 18th century, with onset industrialisation modern agriculture (Krausmann & Haberl, 2002; Lambin Geist, 2006; Ulloa-Torrealba 2020). Ecosystems their contribute human wellbeing functioning societal subsystems (IPBES, 2005) various ways. There conclusive evidence that harmful ecosystem Schmid 2009) consequently nature's contributions people (NCP; Díaz 2015, 2018), as anthropological side entangled nature–culture (Barad, 2007; Haraway, 2016; Haraway Begelke, 2003). Understanding evaluating NCP critical facilitating governments’ decision making if they are achieve a healthy sustainable future (Díaz 2015; Pascual 2017). However, assessing forms take challenging, because majority difficult quantify (Daily, 2000; Daniel Martinez-Alier, 2002). Substantial progress has been made assess material using new data-acquisition tools modelling approaches (de Araujo Barbosa Maes 2015) monetary values (Lautenbach Schmidt Sumarga 2015). non-material contributions—including recreation education, well cultural religious uses, but also aesthetic value appreciation acknowledging nature necessary complement culture—remain measure cannot comprehensively reliably reduced similar single 2017; Seppelt 2011). Fundamental understanding contributions, especially towards wellbeing, quantifying people's valuation biodiversity. Recent attempts include surveys (Ainscough 2019), qualitative quantitative assessments disconnection from products (Celis-Diez Kesebir Kesebir, Prévot-Julliard analyses knowledge graphical media (Wolff The recently established field conservation culturomics further enhances our by analysing word frequencies limited lists mainly contemporary corpora (e.g. social platforms). It does so approximate ongoing change (Michel 2011) means comprehending eventually predicting public interest certain species, areas human–nature interactions (Ladle Willemen Despite these efforts, there still fundamental lack regarding influence on aspects culture its development. Culture seen interpersonally transferred information (Mesoudi, requires communication exist function, subject which far received little attention research NCP. certainly influenced appearance behaviour animals plants semiotics (Tüür Tønnessen, 2014), floriography (language flowers) 19th where plant taxa were related personal character traits (Gagliano Greenaway, 1884), or naming precise colours after animals, ‘violet’ ‘vermilion’ (from Latin ‘worm’) respectively. This biodiversity-aided precision sophistication may development socially technically advanced civilisation, thus facilitates high standard wellbeing. To address lack, we present an approach extracts creative literature, here defined category combining works fiction (constituting preponderance corpus), travelogues, biographies letters. We recognise literature one important form long-preserved allows us degree uses labels biological over time. As texts time, assume usage taxon those correlated awareness part contribution component people. order investigate diachronic development, analysed corpus 16,000 digitised literary covering nearly 300 years English, including English translations originally published other languages. contains more than 1.2 billion tokens (words) searched about 240,000 labels. contrast studies (Kesebir McCrindle Odendaal, 1994; Queiroz Wolff 1999) either only small fraction size set (an between 1 10² labels), aimed comprehensive investigation tried find every non-human living being mentioned corpus. comprehensiveness was precondition calculating diversity both richness (number labels) indices, example Shannon (Magurran McGill, 2011, Chapter 5), (BiL) against background changes general lexical richness, determine number types (unique tokens). allowed calculate several facets borrowing ecological theory 7). By distinguishing different scales (local α-diversity ‘plot scale’, i.e. book section 1,000 words, γ-diversity ‘regions’, size-normalised work), able sequential dissimilarity (β-diversity), is, used sections book. Consequently, how pervasive or, contrast, exceptional use is. acknowledge distribution likely controlled complex drivers percentage authors socialised urban environments, changing poetological/narrotological norms times cultures historical function literature) currently quantitatively. For reason, have deliberately avoided disentangling causal relationships stage. Furthermore, BiL not reflect authors’ awareness, driven processes language transformation, such streamlining vocabulary. Alongside description temporal trends components BiL, nevertheless put forward hypothetical scenario discuss findings fashion: expect growing nature, induced industrialisation, urbanisation extensive industrial forestry century (Brown Harrison, 1978, 2; Grigg, 1987; Cumming, 2016), temporally decrease end century. With time series starting early 1700s, hypothesise initially increases during enlightenment, promoted natural sciences educational system, romanticism, partly interpreted proto-ecological countermovement opposing life (Trepl, 1987), begins understand system interrelated interdependent dynamic elements (Detering, 2020, pp. 307–370; Morton, Rigby, 2014, 2020), reaching peak 1830s. process obtaining data involves steps, illustrate Figure detail following sections. reflecting dynamics 2016) contain sufficiently proportion enable draw random subsamples throughout span analysed, case, 1705–1969. chose period three reasons: (a) provides sufficient print building corpus; (b) point predates agricultural revolution study potential effect time; (c) endpoint digital fundamentally changed access consequences label literature. Subsamples large enough represent population given period, thereby avoiding biases introduced idiosyncrasies attitudes, regions, genres text types. most available corpora, underrepresented biased few canonical works. (i.e. non-selective) content, open access, benefits reproducibility project, working Project Gutenberg. downloaded Standardized Gutenberg Corpus (SPGC; Gerlach Font-Clos, 2020) February 2019. complete SPGC contained approximately 59,000 file basic metadata provided uploaders Library. do works’ publication. Therefore, applied estimation dates birth death corresponding formula was: , central year age 21 death. parameters derived screening sensible ratios each combination, determined deviation actual publication, based randomly gathered subset 4,705 attained lowest 6.9 ratio 0.5. From works, selected all estimated publication date 1705 1969. excluded items categorised text, flags specific keywords ‘fiction’, ‘novel’ ‘travel’, see Appendix A1) included unambiguously ascribed author, opposed institutions departments, universities journals, A2). length individual vocabulary unique tokens. subsampling bias reduction, 15,000 words allow representative described Section 2.4. final 15,798 3,832 authors. Plotting shows uneven distribution, shown 2. reflects status digitisation books generally low publishing prior corresponds strong increase regions corpus, Europe North America, investigated (Biraben, Additionally, first sees successive prominent middle class main producers (Hudson, decline 20th availability domain, turn depends copyright expiration dates. United States, Copyright Term Extension Act 1998 determines work typically enters domain 95 cases depend author. Further preparation required database non-inflected base (lemma) (see 2.2). overcome Lemmatiser Stanford CoreNLP Toolkit (Manning 2014) normalise lemma. step search algorithm inflected aims largest possible meaningful indexes below). bases encyclopaedias Wikispecies Wikidata (Vrandečić Krötzsch, sources, collected taxonomic data. Subsequently, extracted compiled labels, call could rule out represented spelling variants addition hand, would result replacing spaces hyphens vice versa omitting spaces, apostrophes, leading adjectives altogether. alphabet simplifying letters diacritics), retaining apostrophes additional symbols. manually produced blacklist (Appendix A3) containing ambiguous probability generating false positives homography ‘bishop’, ‘diver’ ‘ray’) artefacts extraction ‘european’, ‘alexander’ ‘red’), indistinguishable formatting original match list token sequences characters identical. why automated extraction, particular when concerning unusual rarely names, dependent accuracy consistent order, parentheses correct orthography) providing via openly encyclopaedias. During label, findings, like adjectives, locations blacklist. resulting entries domains levels. After above 161,488 Wikispecies, added another 80,955 Wikidata, totalling 242,443 35,588 synonyms 106,592 variants. Altogether, refer 100,263 taxa; 214,941 91,244 level. get quantification consistency database, same manner database. preserved characters: lines, hyphens, full stops, exclamation marks, question commas, semicolons, quotation marks numbers. Each 1,559,771 frames When occurrence entry generated comprising scientific name, frame number. occurrences BiL. analysis grouped into intervals 5 years, according 1705. interval, 10 If fewer 5-year them prone chance extreme measurement, repeated hundred interval averaged parameters, introduce below. plot some predetermined area typical investigations 2). regard α-region determined. measures species-area ecology, normalisation selecting work. Again, iterated average parameters. normalised referred γ-region characterised γ-diversity. abundance cross-comparison, β-diversity exact Table 1. avoid leave space selection, 15 (corresponding words). iteration through (1 → S) p respective dissimilarity, Whittaker (1960), reduce decreasing variety obtain proxy compared (as above) overall terms distinguished without separating taxon, periods. should (counting separately) taxa. whole synonymy among period. size-related sampled 100 frames, each, measurements repeatedly times. More two thirds least (1,066,839 frames; 3). total, revealed 4,416,187 5,994 4,652 One third (1,778,885 4,416,187) identified level (3,076 4,652). significant difference (one occurrences) evidently known (two taxa). Instead folk-biological (Atran Medin, 2008), organisms presented generic level, roughly equating genus family ‘oak’, ‘eagle’ ‘deer’, referring locally dominant ‘European oak’, ‘golden eagle’ ‘red deer’ Similar rank curve 4 abundant representing tendency life-form 1997) higher, ‘tree’, ‘bird’ ‘animal’, beings humans frequently interact with, domesticated plants, ‘horse’, ‘rose’ ‘coffee’, pose threats, ‘bear’, ‘wolf’ ‘lion’. analogue relationships, expected saturation fitting Michaelis–Menten curve, half-saturation constant 139,000 190 (Figure 5). Although appears scattered, fit highly (p < 0.001). differ substantially œuvre. Generally, exceptionally (40,000+) content less 1%. author highest 19th-century novelist Charlotte Mary Yonge 903 6 illustrates relationship total indicating mean 1.09 ± 0.46% across Variation high, ranging 0.37% 2.22% (percentiles 2.5 97.5 respectively). writing under study, estimates abundance, observed clear 1830s measures, fitted regression model graph. reaches 1835 then decreases 7, top left; R² = 0.59, slightly trend right): maximum 1836, remains almost until around 1955 before it abruptly, slight (R² 0.58, Combining results (abundance), did maintain (richness), induces increasing redundancy 1835. upper centre left), quantified diversity, falls line supports richness. increase, reached 1837 decreased γ-diversity, right) (1960) dividing per α-diversity. Here, inverse comparison initial 1831 0.55, At 1835, maximum, minimum uniformly distributed sets things 1,830 1950, meaning former similarity course work, while dwindled, indicated synchronous left) represents ignoring correlation synonyms. lower left, show distinct taxa, followed 1805, decades 0.46, comparison, underwent steeper later. conflation consequence streamlining, calculated right. case yields insignificant 0.16, 0.31), implying largely stable count compare bottom labels’ right), observe relative increased 1832 afterwards 0.63, 0.01). richness: 0.57, stronger beings, conduct versions. 1830s, shed light causes explored unrelated awareness: systematic streamlining. declined faster indicates used. accept patterns serve indicators make probable. stated progressively disappear declining exposure (and authors) daily lives intensified land (Seppelt 2016). though driver variables quantifying, raised cities claim relationship, historic co-occurrence industrialisation/urbanisation. will subsequent might correspond scenario. Finally, limits approach, relation reasons exhibit coverage includes thematic formal emergence rise novel paradigm (Davis, 1983; Watt, 1957). beginning therefore rather low. want highlight Google Books (M. Davies, 2011), albeit overwhelming extent, zero five annually 1,700 1,720 even before. confirms comparatively all, current remainder primarily mostly confirm religious, moral, political norms. emancipation normative successively developed broader representation increasingly covered topics surrounding environment.