Rethinking Historical Method in the Digital Age: Text Mining, Geographical Information System, and Network Analysis

1. Introduction

The rapid growth of digitised archives has transformed historical research by changing the scale and type of evidence accessible to historians.Oftentimes, scholars are faced with a glut of newspapers, correspondence files, maps, census data, and born-digital records that cannot be handled by reading them by hand. In this shifting terrain, digital history tools help historians define evidence, articulate definitions of it, and communicate knowledge [1].With this change the importance of computational methods has increased. Historians can use text mining to study large collections of texts, looking for recurring themes, shifts in concepts, and patterns of discourse that are difficult to detect through close reading alone[2].GIS enables scholars to examine movement, settlement, trade, the environment, and territorial change as spatial phenomena rather than merely descriptive narratives [3].

Network analysis reconstructs relationships between people, exposing circulation, brokering and association, that is otherwise hidden. Historian insist Digital methods and tools are not of any value, if they are not used with interpretative restraint. Selection, preservation, cataloging and access all connect to digitized archives.Therefore, online content is not a neutral representation of the past[4].Textual corpora may contain OCR(Optical Character Recognition) errors, missing metadata and uneven digitisation. If computational outputs are taken as facts, rather than as tentative prompts for analysis, these problems can skew results[5].The case is supported by evidence from historical GIS research. GIS is more than digital mapping; it’s a spatial database and analytical framework. It links each item of information with a particular place, enabling historians to systematically link and visualise geographically relevant evidence [6]. There are issues with historical geographical data: place names change, boundaries move and historical locations don’t always match up with modern coordinates. Therefore, historical context and critical source analysis must be intimately link spatial visualisation[7].

Network analysis can enrich historical research, especially in the areas of influence, exchange, kinship, intellectual connections, or institutional ties. But the seeming precision of a network graph can mask the patchy and inconsistent nature of the underlying records, necessitating careful interpretation at every stage of the analysis, noted Wetherell[8]. The importance of computational methods in history is not to replace traditional historical practices but to enhance them. The best way to practice digital history is with accuracy, computational investigation with contextual comprehension, and technical innovation with historiographical insight[9]. This manuscript builds on that premise and contends that text mining, GIS, and network analysis should be viewed as valuable enhancements to the historian’s toolkit rather than as a departure from the discipline’s interpretive foundations.

• Test Mining and Analysis of Historical Corpora

In digital history, text mining has emerged as one of the most influential computational techniques, as it enables historians to analyse extensive textual collections that are either too dispersed, too repetitive, or too vast for conventional reading. The growing number of digitised sources havehaschanged both the amount of information available to historians and the types of questions they can ask about language, discourse, and change over time, as noted by scholars of digital history[10]. This is particularly true where historical interpretation is concerned with tracking a theme, a conceptual change or a rhetorical pattern across thousands of pages of newspapers, letters, parliamentary debates, court proceedings, and pamphlets, which no individual researcher could reasonably read from beginning to end[11].As for methods, text mining involves keyword analysis, collocation analysis, topic modelling, semantic clustering and named-entity extraction. While these methods do not automatically generate history, they can help historians to find patterns in large corpora which might otherwise go unnoticed. For example, Grajzl and Murrel show that topic modelling can be a useful tool for legal-historical researchers to create general quantitative narratives from a large body of case law and then use those findings to ask historically relevant questions about continuity and change[12].The value of their work is that it shows how computational analysis can inform historical interpretation when the method is applied with care and the results are interpreted in terms of substantive historical questions, not as ends in themselves[13].

Text mining is particularly useful in research using digitised newspaper archives where the task is not only to collect text but also to construct a coherent and usable corpus. Joshua Wilson Black advocates for a more careful and iterative approach to creating specialised corpora from digitised collections. Simple keyword searches can often be inadequate for working with large newspaper archives.They can remove content from context, and they can produce highly uneven results, leading to potential misinterpretations of historical narratives and inaccuracies in the analysis of trends over time[14].This awareness is relevant for historical text mining, as before applying any analytical model, the corpus needs to be collected, cleaned and delimited to reach high-quality results[15].In other words, text mining is about corpus generation,not the algorithm, and that is a kind of historical judgement[16].But historians who use digital technologies often point out that historical textual material is unreliable and often misleading if not handled with care. Digital archives, before they even interpreted: OCR problems, irregular spelling, damaged originals, incomplete newspaper runs, poor metadata and selective digitisation. Guldi stresses that historians should think critically about the relationship between data, algorithms, secondary sources and interpretations rather than assuming that a computational solution offers a clear view of the past[17].Her conception of ‘critical search’ is particularly apt here, as it requires digital history to evolve through recursive verification, methodological fit and interpretative self-awareness, than technological confidence alone[18].

This warning also reflects a more general concern in the historiography. Lara Putnam has convincingly argued that searchable digital archives cast “shadows” that shape the historian’s gaze, tending to focus on what is already digitised, searchable, and linguistically legible while masking absences and distortions in the archive [19].This means that text mining should not be viewed as a neutral discovery method. Instead, it should be used as an interpretative and exploratory tool to help researchers move between close reading, source-based explanation, and large-scale pattern recognition [20].This is why the value of text mining to historical research is more based on augmentation than on automation. It helps the historian to identify trends, test impressions, sharpen questions, and return to sources with greater conceptual clarity. When used correctly, it enhances the possibilities for historical research without compromising the core values of the discipline: context, evidence, and critical interpretation[21].

• GIS and Spatial History

Historical GIS enables connecting the historical evidence with place, size and movement. In this way, spatial relations, which go beyond the narration itself, can be used for understanding the study of boundaries, settlements, movements, trade, health issues and environmental change. GIS has become a significant tool in historical study because it permits historians to approach geography not as a backdrop but as an analytical feature of the past[22].This move is important because of the uneven character of historical processes in various territories. GIS allows historians to create georeferenced maps, to link archival documents to locations, and to visualise change over time in ways difficult to convey through words alone. A recent historical GIS study of the nineteenth and early twentieth centuries mapped inhabited alleys and linked spatial data to public health and housing records. These developments prompted new insights into segregation, disease anxieties, and racialised reform[23]. Likewise, a historical GIS study on missionary education in Ottoman Turkey used spatially referenced data to trace the growth of women’s education that might add depth to historical insights about regional variation and institutional diffusion [24].Previous academic research on GIS has also indicated that GIS is more than digital mapping. It is a spatial database and analytical framework, integrating and analysing the locations, attributes and temporal layers as a whole. Recent scholarship in the history of maps has shown that digital mapping has both increased the use of historical maps as sources and improved the production of maps as historical arguments. This kind of development enables historians to synthesise evidence and create more dynamic interpretations of urban and regional change [25].That’s important because digital historical maps don’t just reflect where things went, they often actively help to create those conclusions[26].

Historians have found GIS especially helpful in trying to understand complicated relationships between culture and geography. Recent developments in the spatial humanities and historical geography have demonstrated that geospatial tools can reveal underlying structures in urban development, resource distribution, route systems, and cultural landscapes by visualising and comparing spatial relationships across datasets[27].Moreover, new digital techniques, such as symbol detection, map vectorisation and improved place-name searches, allow the extraction and interpretationof information from historical maps, thus expanding the pool of evidence for GIS-based historical analysis[28].At the same time, historians are sounding the alarm about the need for interpretative care in GIS-based research. Historical places do not always match up with modern places. Boundaries change, names change, and many places are known only in vague or incomplete form. Current debates in spatial history suggest that maps and visualisations shape assumptions about causality, authority and certainty. Thus, the spatial output should be critically analysed and not viewed as a clear representation of the past[29].

Even the most technically sophisticated GIS operations still rely on judgements of georeferencing, categorisation, size and omission, all of which affect the final arguments [30].While historians are still held accountable for source critique, uncertainty and context, GIS used properly allows them to move away from description towards geographical explanation[31].

• Network Analysis and Historical Relationships

Network analysis is increasingly useful in the study of history because it allows researchers to examine links between people and not just a single person. Movement patterns can be recreated using network approaches, brokering connections and influence between merchants, intellectuals, officials, families, activists and organisations [32].Instead of focusing solely on major events or renowned individuals, historians can gain deeper insights. In the words of Aleksandra Kaye, social network analysis can help historians to build ‘big pictures’ by revealing connections and structures that are often missed by traditional centre-periphery narrative or ‘great man’ approaches [33].This is especially important when questions of mobility, exchange or relational power arise in historical contexts. Recent network-based studies have shown that the importance of historical persons or places is not always apparent by virtue of their prominence in texts. what seems marginal in narrative sources can suddenly appear crucial when their connective role is investigated systematically. Network analysis can produce alternative views on the emergence of communities, the circulation of knowledge, and the importance of structures in historical systems [34].This flexibility is also an advantage, as network analysis can be applied to fragmented historical material. Historians often have to reconstruct relationships from letters, institutional records, membership lists, travel traces or printed references, rather than from complete relational datasets. Recent mixed-methods digital scholarship in the humanities shows how network analysis can be combined with controversy mapping and traditional humanities approaches to study eighteenth-century debates. This means that digital relational approaches are best used in combination with, rather than as a replacement for, close reading[35].But historians always warn caution in interpretation. Kaye contends that digital methods demand considerable expertise in reading and communicating their results; graphs and measurements are not self-evident and can be deceptive when removed from their historical context[36]. In this way, Tidman’s work points out that network visualisations are not independent demonstrations of historical truth but rather analytical tools within a large mixed-method framework[37].Such considerations are particularly needed as historical archives may not preserve relationships uniformly. Ambiguous links, missing actors, uncertain dates, and selective survival can distort the apparent structure of a network. Historians need to reassess prominence, reveal hidden intermediaries, and trace the historical development of communities, using network analysis as an interpretive framework. But they must use network analyses in accordance with evidence, context and source criticism [38].

• Methodological Limits and Interpretive Caution

The growing application of computational methods in history has necessitated a methodological caution. Digital archives may look comprehensive, but historians warn that digitisation is patchy and selective. This selectivity means that the searchable record is often a reflection of institutional priorities, funding decisions, cataloguing practices, and preservation politics rather than of the complete surviving record[39].Computational history begins with an archive, not neutral data. One of the most important limitations is the bias of sources and the representativeness problem. Beelen and his co-authors show that digitised newspaper collections may contain hidden distortions in coverage, regional balance and preservation. Such biases can lead researchers to assume that large datasets are inherently representative of past print culture[40].Digitisation tends to prefer content that is legible, catalogued and machine-readable already, while obscuring the inherent absences in archival survival itself. For this reason, according to Putnam, searchable archives are a ‘shadow’themselves[41].The second limitation concerns the OCR errors and instability of the text. Jo Guldi notes that textual data from the past can be ‘dangerous’ because OCR systems, particularly when applied to older print cultures, damaged originals, or non-Western scripts, have the potential to introduce systematic distortions into the corpus before interpretation[42].Recent digital global history literature stresses that OCR error rates are high, particularly in less expensive newspapers and mass publications, which are also unevenly represented in digital collections[43].When the full text is available, computational results may be more indicative of the quality of the scanning and text extraction than of the historical phenomenon being investigated. A third worry is the quality of the metadata and the construction of the corpus. The effectiveness of AI and large-scale digital archival workflows hinges on the quality of metadata, institutional trust and collaborative curation, according to Jaillant. Therefore, weak metadata hampers impact discovery and classification and, subsequently, interpretation[44].Joshua Wilson Black shows that the construction of a historical corpus is not simply a technical operation but a scholarly act of delimitation. Patterns in text mining are influenced by the selection, cleaning and grouping of texts, which can suggestively affect the outcomes and interpretations of the analysis conducted on historical corpora[45].

In their work with space historians are confronted with the problem of ambiguous georeferencing. GIS outputs may involve interpretive decisions about fit, scale, and placement because historical places change, boundaries shift, and many historical sites do not translate neatly into present-day coordinates. Georeferencing accuracy is not just a technical issue, as current geospatial research suggests[46].Any historical argument that depends on spatial visualisation or layered mapping, is affected by its accuracy. Maps created using GIS should thus not be considered clear depictions of past geography but rather as analytical reconstructions. Network-based research is also plagued by the issues of unstable entity matching and missing data. Historical archives rarely preserve relationships in their entirety. When some actors or documents survive in disproportionate numbers, the selection and availability bias of digital archives can seriously distort correspondence networks or institutional ties[47].More generally, recent survey work on network data suggests that missing data, if not carefully handled, can bias inference and result in unreliable analytical results[48].The problem is even more acute when it comes to historical research, when names are ambiguous, dates are ambiguous, and identities are lost in multiple spellings or systems of archives. These difficulties point to a larger danger: the over-quantification of shaky historical data. Lang and his collaborators suggest that a too-close association with a culture of programming and quantitative bias risks marginalising issues that require ambiguity, contradiction and interpretative openness rather than just precise measurement[49].Thus, the most powerful digital history computational output is not taken as impartial truth. Indeed, it sees models, maps, graphs, and clusters as one layer of evidence to be judged against provenance, context, archival circumstances and historical judgements[50].In this sense, interpretative care is an aspect of the approach per se rather than a binding of the external computational history. Digital technologies are expanding research, but their results are only persuasive if historians remain sensitive to bias, incompleteness, uncertainty, and the politics of data generation and preservation which can knowingly affect the interpretation and reliability of historical narratives [51].

• Conclusion

The digital revolution in historical research has not replaced traditional historical methods but has enormously increased the historian’s toolkit and has reshaped the possible scope of historical inquiry. We are able to interrogate sources at scales and in configurations previously unimaginable using text mining, GIS and network analysis- identifying patterns across thousands of documents, visualising spatial phenomena, and reconstructing networks of individuals and institutions[52].They provide new avenues for examining old questions and new avenues for examining change, continuity and connection within the historical record.Thus, the article argues that these computational methods are not merely technical add-ons, but tools that require critical engagement and reflexivity. Digital outputs have long been cautioned against as transparent windows onto the past by scholars. Selection bias, OCR errors, and the quality of the metadata are permanent problems that subtly but significantly affect both the corpus and the results[53].Spatial and network analyses can be visually very compelling but need to be understood in terms of changing boundaries, incomplete records, and the uncertainties inherent in historical data [54].The computational results are only valid when combined with archival thoroughness, contextual understanding and interpretive prudence, as has been forcefully argued in recent scholarship[55].

The future of historical research will depend on how well we can meld the strengths of the digital and the traditional. As digital archives grow and AI-assisted tools become ubiquitous, historians must be cognisant of the politics of digitisation, the infrastructures that govern access, and the collaborative ethics that underlie rigorous and trustworthy research[56]. In the end, digital history promises not to replace established historical practice but to amplify it, making possible more comparative, reflexive, and innovative research while preserving the discipline’s core values of evidence, accountability, and interpretive nuance [57]. Only by such integration can the opportunities of the digital age be fully realised by historical scholarship.

References

[1] Lassig, Simone. “Digital History.” Geschichte und Gesellschaft, Vol. 47 (1), 2021, pp. 5–8; Zaagsma, Gerben. “Digital History and the Politics of Digitization.” Digital Scholarship in the Humanities 38 (2), 2023, pp. 831–835.

[2] Putnam, Lara. “The Transnational and the Text-Searchable: Digitized Sources and the Shadows They Cast.” American Historical Review 121 (2), 2016, pp. 377-381.

[3] DeBats, Donald A., and Ian N. Gregory. “Introduction to Historical GIS and the Study of Urban History.” Social Science History 35 (4), 2011, pp. 455–458.

[4] Zaagsma. “Digital History and the Politics of Digitization.” pp. 831–835; Peucker, Paul. “The Role of the Archives in a Digital World.” Journal of Moravian History 22 (2), 2022, pp. 157–162

[5] Guldi, Jo. The Dangerous Art of Text Mining: A Methodology for Digital History. Cambridge: Cambridge University Press. 2023, pp. 1-8.

[6] Gregory, Ian N., and Paul S. Ell. Historical GIS: Technologies, Methodologies and Scholarship. Cambridge: Cambridge University Press, 2007, pp. 1-20; DeBatsand Gregory. “Introduction to Historical GIS”, pp. 455–458

[7] Davis, Martin. “Digital Technology and the Future of Map History.” Imago Mundi: The International Journal for the History of Cartography, 77 (2), 2025, pp. 221–225.

[8] Wetherell. “Historical Social Network Analysis”, pp. 125–128.

[9] Lassig. “Digital History”, pp. 27-34; Zaagsma, Gerben. “Digital History and the Politics of Digitization.” Digital Scholarship in the Humanities, pp. 842–849.

[10] Putnam. “The Transnational and the Text-Searchable”, pp.377-381; Guldi. The Dangerous Art of Text Mining, pp. 25-35.

[11] Putnam. “The Transnational and the Text-Searchable”, pp.377-381.

[12] Grajzl, Peter, and Peter Murrell. “Using Topic-Modeling in Legal History, with an Application to Pre-Industrial English Case Law on Finance.” Law and History Review 40 (2), 2022, pp.189–194.

[13] Ibid., pp. 214-228.

[14] Black, Joshua Wilson. “Creating Specialized Corpora from Digitized Historical Newspaper Archives: An Iterative Bootstrapping Approach.” Digital Scholarship in the Humanities 38 (2), 2023, pp. 778-783.

[15] Ibid., pp. 779-792.

[16] Ibid., pp. 779-783.

[17] Guldi. The Dangerous Art of Text Mining, pp. 25-56.

[18] Ibid., pp. 25-56.

[19] Putnam. “The Transnational and the Text-Searchable”, pp.392-402.

[20] Ibid., pp.392-402; Guldi. The Dangerous Art of Text Mining, pp. 1-22.

[21] Black. “Creating Specialized Corpora from Digitized Historical Newspaper Archives”, pp. 779-783, 791-92; Grajzl and Murrell. “Using Topic-Modeling in Legal History, with an Application to Pre-Industrial English Case Law on Finance”, pp.189–194; Guldi.The Dangerous Art of Text Mining, pp. 25-35.

[22] Colson, Justin. “Layering London’s History: Digital Mapping and Spatial History.” Imago Mundi, 77 (2), 2025, pp. 168–188; van Mil, Yvonne, Carola Hein, and Vincent Baptist. 2025. “The Potential of Geospatial Technologies and Open Data in Planning History.” Planning Perspectives, 41 (1), 2025, pp. 185–298.

[23] Swope, Carolyn B. “The Spatial Configuration of Segregation, Elite Fears of Disease, and Housing Reform in Washington, D.C.’s Inhabited Alleys.” Social Science History 48 (2), 2024, pp. 173–173, 190-199.

[24] Amasyalı, Emre. “Protestant Missionary Education and the Diffusion of Women’s Education in Ottoman Turkey: A Historical GIS Analysis.” Social Science History 46 (1), 2022, pp. 173–222.

[25] Colson. “Layering London’s History: Digital Mapping and Spatial History”, pp. 168-175; Davis. “Digital Technology and the Future of Map History”, pp. 221–23.

[26] Scholz, Luca. “The Atmosphere in Spatial History: Digital Evidence and Visual Argument.” Past & Present 269 (1), 2025, p. 130.

[27] Swope. “The Spatial Configuration of Segregation”, pp. 173–173, 190-199; van Milet al., “The Potential of Geospatial Technologies”, pp. 185- 90.

[28] Pai, Pi-Ling, Hsiang-Hsi Liu, Wen-Rong Su, and others. 2025. “AI-Supported Smart Search for Multi-Period Historical Maps Using the mapKurator Place Name Index.” Journal of Map & Geography Libraries 22 (1), 2025, pp. 1–6; Petitpierre, Rémi, and Pierre Guhennec. “Effective Annotation for the Automatic Vectorization of Cadastral Maps.” Digital Scholarship in the Humanities 38 (3), 2023, pp. 1227–1231.

[29] Scholz. “The Atmosphere in Spatial History”, p. 130; Davis. “Digital Technology and the Future of Map History”, pp. 221–223.

[30] Colson. “Layering London’s History”, pp. 176-182; Davis. “Digital Technology and the Future of Map History”, pp. 221–225.

[31] Swope. “The Spatial Configuration of Segregation”, pp. 190-199; van Milet al., “The Potential of Geospatial Technologies”, pp. 185- 91; Scholz, Luca. “The Atmosphere in Spatial History: Digital Evidence and Visual Argument.” Past & Present 269 (1), 2025, p. 130.

[32] Kaye, Aleksandra. “Piecing Together ‘Big Pictures’ with Social Network Analysis and Digital Tools.” BJHS Themes 9,2024, pp. 149–153.

[33] Ibid., pp. 149–153.

[34] Ibid., pp. 160-167.

[35] Tidman, Gemma, and Marc Sarazin. “Movers and Haters: Dynamic Network Visualisation and Analysis of an Historical French Controversy.” Digital Scholarship in the Humanities, 2025, pp. 1-5, 18-22.

[36] Kaye. “Piecing Together ‘Big Pictures’”, pp. 153-157.

[37] Tidmanand Sarazin. “Movers and Haters”, pp.18-22.

[38] Kaye. “Piecing Together ‘Big Pictures’”, pp. 125-128.

[39] Zaagsma. “Digital History and the Politics of Digitization”, pp. 831–835; Mäkelä, Eetu, James Misson, Devani Singh, Mikko Tolonen, and others. 2026. “Opening the Black Box of EEBO.” Digital Scholarship in the Humanities 41 (1), 2026, pp. 236–240.

[40] Beelen, Kaspar, Jon Lawrence, Daniel C. S. Wilson, and David Beavan. “Bias and Representativeness in Digitized Newspaper Collections: Introducing the Environmental Scan.” Digital Scholarship in the Humanities 38 (1), 2023, pp. 1-3, 18-22; Beelen, Kaspar, Jon Lawrence, Daniel C. S. Wilson, David Beavan, and others. “Whose News? Critical Methods for Assessing Bias in Large Historical Datasets.” Computational Humanities Research 1, 2025, pp. 1-3.

[41] Putnam. “The Transnational and the Text-Searchable”, pp.392-402.

[42] Guldi. The Dangerous Art of Text Mining, pp. 25-35.

[43] Toye, Richard, Astrid Swenson, Haakon A. Ikonomou, and others. “What Is Digital Global History Now?” Transactions of the Royal Historical Society 35, 2025, p. 450.

[44] Jaillant, Lise, and Arran Rees. “Applying AI to Digital Archives: Trust, Collaboration and Shared Professional Ethics.” Digital Scholarship in the Humanities 38 (2), 2023, pp. 571–576.

[45] Black. “Creating Specialized Corpora from Digitized Historical Newspaper Archives”, pp. 779-783, 791-92.

[46] Pope, Rere-No-A-Rangi, and Marcus Frean. “Georeferencing Historical Maps at Scale.” In 13th International Conference on Geographic Information Science (GIScience 2025), LIPIcs 346, 2025, pp. 11:1–11:3.

[47] Buarque, Breno S., Raphael Vogl, and Aleksandra Kaye. “An Agent-Based Model to Build Letter Correspondence Networks.” Digital Scholarship in the Humanities 40 (4), 2025, pp. 1101–1104.

[48] Shaillanto, Q., and others. “A Survey on Missing Data Generation in Networks.” Mathematics 14 (2), 2026, pp. 341-344.

[49] Lang, Sarah A., and coauthors. “(Doing) Computational History: The Role of Data Work in a Field in Formation.” Journal of Open Humanities Data 6 (2), 2026, pp. 26:1-26:4.

[50] Zaagsma. “Digital History and the Politics of Digitization”, pp. 842–849; Jaillant and Rees. “Applying AI to Digital Archives”, pp. 571–576; Guldi.The Dangerous Art of Text Mining, pp. 25-35.

[51] Mäkeläet al., “Opening the Black Box of EEBO”, pp. 236–240; Beelenet al., “Whose News? Critical Methods for Assessing Bias in Large Historical Datasets”, pp. 1-3.

[52] Putnam. “The Transnational and the Text-Searchable”, pp. 377-381;Swope. “The Spatial Configuration of Segregation”, pp. 190-199; Kaye. “Piecing Together ‘Big Pictures’”, pp. 149-153.

[53] Guldi. The Dangerous Art of Text Mining, pp. 25-35; Beelenet al., “Bias and Representativeness in Digitized Newspaper Collections”, pp. 1-3; Black. “Creating Specialized Corpora from Digitized Historical Newspaper Archives”, pp. 779-792.

[54] Davis. “Digital Technology and the Future of Map History”, pp. 221–223; Tidmanand Sarazin. “Movers and Haters”, pp. 1-5, 18-22.

[55] Lassig. “Digital History”, pp. 27-34; Zaagsma. “Digital History and the Politics of Digitization”, pp. 842–849.

[56] Jaillant and Rees. “Applying AI to Digital Archives”, pp. 571–576; Zaagsma. “Digital History and the Politics of Digitization”, pp. 831-835.

[57] Toyeet al., “What Is Digital Global History Now?”, pp. 445-447; Guldi.The Dangerous Art of Text Mining, pp. 25-35.