Keywords

1 Introduction

The Amsterdam Time Machine (ATM) aims to provide an integrated platform for presenting historical information about people, places, relations, events, and objects in a spatial and temporal context, with a focus on the city of Amsterdam. The web of data on the history of Amsterdam is created by systematically linking existing datasets from social science and humanities research with municipal and cultural heritage data, where possible in the form of Linked Open Data. The linked data can then be organised and presented in spatial representations, such as geographical and 3D visualisations. The result is a virtual replica of the city, which allows users to explore the city through space and time, at the level of neighbourhoods, streets, or individual houses.

Recently, the authors collaborated in developing a first proof of concept that connects linked data from the Amsterdam cultural heritage institutions and various scholarly research projects to a GIS infrastructure that provides the historical geographical and topological context for these linked datasets, focusing on the period 1832–1921.Footnote 1 Such an infrastructure has to be based on precisely identified and localised historical addresses, since these function as the key to a large amount of civil and fiscal data. The Fryske Akademy HisGIS (Historical GIS) team that developed the GIS infrastructure introduced the principle of geographical coordinates as anchor points for all historical data with a spatial component that cannot be tied to specific geometries of buildings or plots. The location points prevent the most common pitfalls of linking historical addresses with specific geometries of parcels or buildings as starting points, which because of changes through time eventually may lead to fuzzy and inaccurate connections caused by historical mutations such as the merging, demolition, split or aggregation of houses and their addresses.

In addition to building the GIS infrastructure, the aim was to test the use of such a geo-spatial platform as an instrument for digital humanities research. Building on the concept of ‘deep maps’ [3] - geographical representations of data on both the material and experiential dimensions of a place - the Amsterdam Time Machine geo-spatial platform aims to facilitate ‘scalable digital humanities research’: smoothly navigating historical data from the micro level of one location, anecdote or document to the macro level of patterns in large, linked datasets that expose broader social and cultural processes. The Amsterdam Time Machine operationalises this by investigating the urban history of Amsterdam on a scale that varies between the micro level of a plot, person or place and the macro level of broader societal processes in the city as a whole - a microscope and telescope in one [12, 44]. Such research environments offer an unprecedented opportunity to explore the relationship between physical and social space and how this connection was experienced and transformed over time.

To test the potential of the ATM platform as a research instrument, the authors conducted a small, pilot use case on Amsterdam entertainment culture in the late 19th and early 20th century. The use case brought together data and insights from three different humanities disciplines - linguistics, socio-economic history, and media studies - to study the emergence of cinema as a new form of entertainment at the turn of the twentieth century against the background of the existing forms of entertainment and the socio-economic composition of the neighbourhoods in which this entertainment was located. Combining data on Amsterdam dialects, occupational status and leisure venues and visualising them in the GIS infrastructure allowed us to test the platform’s capacity for interdisciplinary research, by making a connection between social, economic and cultural dimensions of turn of the 20th century urban life in the capital.

2 Related Work

The core idea of the Amsterdam Time Machine revolves around anchoring information to time and place. This concept is not novel, but its scale as envisioned by the European Time Machine [48] is. The first project in this consortium to present this grand vision was the Venice Time Machine [23], which showcased a walk through the Venice of 1,000 years ago via digitised and transcribed maps, registries and images. The current constellation of Time Machine projects has been made possible by the groundwork laid by sub-disciplines such as spatial humanities [2, 3] and projects such as the Reassembling the Republic of Letters COST Action (2015–2018).Footnote 2

A relatively new development is the cross-over between humanities and semantic web research. Projects such as FDR/Pearl Harbor [22], Agora [50], Dutch Ships and Sailors [8], and the Semantic Sampo portals [21] are exploring the possibilities of representing and connecting concepts from historical documents for advanced search and analysis. The advantage of modelling concepts according to semantic web standards, is that they use a uniform format and when modelled using shared ontologies, can be connected to other sources in the Linked Open Data cloud [33].

The Amsterdam Time Machine project currently focuses on three use cases: linguistics, media studies, and socio-economic history, which are connected through a geotemporal link. Due to its historical focus on the city of Amsterdam, the project is also positioned in the realm of urban history. As this is not a survey paper, we briefly highlight work on the intersection of semantic web and each of these domains. For a broader overview of semantic web technologies in the humanities, the reader is referred to [35].

The linguistics community has been computationally modelling language for over 4 decades, with WordNet as its most famous lexical database [39]. Semantic web-ready models such as the OntoLex-Lemon [34] have evolved into W3C backed standards. Since 2011, the community is working towards keeping track of semantic web datasets describing linguistic phenomena in the linguistic linked open data cloud [5].

Many historians in the field of social and economic history work with structured or tabular data such as census data. This type of data lends itself well to integration with the semantic web. In the Netherlands, the CEDAR project [36] and follow-up use cases within CLARIAH-CORE and CLARIAH-PLUSFootnote 3 have shown that converting and integrating social and economic data can ease scholarly workflows and lead to new insights [19, 37].

Digitised media have been the topic of interest in many semantic web research projects (cf, [46], Multimedian) but often with a focus on improving data annotation and object retrieval. Some work has attempted to create links between media, location and the past (cf. [17]) but to the best of our knowledge, the Amsterdam Time Machine project is the first project that aims to leverage semantic web standards to answer humanities scholars’ urban history and media studies research questions.

3 Semantic Deep Mapping

3.1 Deep Mapping as Concept and Method for Researching Urban History

The focus on space as an angle from which to explore people, places and events in the history of a city is not new. As historian Jo Guldi has explained, the interest of humanities researchers in space as an angle for analysis dates back to the 1880s, “when scholars in history, religion, and psychology reflected on our nature as beings situated in space.” [15] From the 1970s, under the influence of French philosophers such as Michel Foucault, Henri Lefebvre, Michel de Certeau and Paul Virilio, the analysis of space has been connected to questions of power, causing a true ‘spatial turn’ in various humanities disciplines. For urban historians, this meant that they considered the city with a “renewed interest” in the ways in which the “microcosms of everyday life” are related to the “macrocosms of global flows”. [15] Or, in the words of historian Charles Tilly, how they rediscovered the city as a “privileged site for study of the interaction between large social processes and routines of local life” [49, p. 704].

This revived interest in space as a core angle for the analysis of everyday local life was greatly aided by the emergence of new, digital technology, in particular the Geographical Information System (GIS) developed in the 1960s by the Canada Land Inventory. GIS allowed scholars in the humanities and social sciences to create digital maps on which data on various phenomena could be visualised and analysed at different levels of aggregation. Within the humanities, GIS has been most routinely employed by archaeologists for organizing and analyzing their excavation data. After that, GIS has also been discovered in other areas, such as Literary Studies [26], Film Studies [20] and History [14, 32]. As David Bodenhamer, John Corrigan and Trevor M. Harris point out, a cartographical representation of historical data “provides fresh perspective and new insights into the study of culture and society.” [3, p. 2]

In an attempt to do justice to the complexity of human culture that is the topic of humanities research, Bodenhamer, Corrigan and Harris have proposed the concept of ‘deep maps’ as the next step in humanities research [3]. A deep map extends the focus of geographical information systems on tangible or material aspects of space with attention for the way people attribute meaning to specific places. [3, p. 3] This takes the form of adding sources that document the use and experience of public space at a given moment and as it evolved over time. Such a ‘thickening’ of the information layers included in GISs eventually allows for the realisation of ‘mirror worlds’: virtual places that correspond to actual places [11, 24]. It is the ambition of the European Time Machine project to leverage Europe’s digital cultural heritage for extending such mirror worlds with the dimension of time, so that we can explore the rich complexity of human culture from both a spatial and temporal perspective.Footnote 4

Various scholars in literature, history, archeology and media studies have experimented with the concept of deep mapping, enriching cartographic representations of specific places with sources that document the use and experience of the place, for example, as imagined by artists (e.g., [30]) or figuring in popular movies and television series (e.g., [4]). Our approach aims to expand upon such existing work by creating an infrastructure with stable location points and a Linked Open Data approach that allows for deep mapping at scale. Such an approach, which harmonises various types of data sources on specific places across different times in one GIS infrastructure, provides the foundation of the mirror world foreseen in the Time Machine project.

3.2 GIS Infrastructure

Providing a solid GIS infrastructure for the Amsterdam Time Machine presented us with various challenges. Modelling the data of the use cases into a uniform and spatial accurate system starts with a semantic discussion. This requires an unambiguous explanation and use of seemingly obvious concepts such as ‘houses’, ‘buildings’ and ‘addresses’. These elements function as a key to a large amount of civil and fiscal data, given the fact that until late in the twentieth century, citizens were registered with their address instead of a personal identification number. Therefore, censuses, resident registrations and fiscal administrations contain a mass of information on people in the past. Together, they provide a ‘collective’ identification of the people of Amsterdam.

However, in many (or most) cases the present-day address is not the same as a historical address. It often happens that even the physical building is no longer the same. It should also be kept in mind that addresses were originally introduced for fiscal, military or administrative purposes, not as a convenient system for the general public to find the right houses [41, 47]. Over the course of the 18th and 19th centuries, the system of house registration was changed several times. The first visually shown numbers on Amsterdam facades originate from 1795; the system then was organised by district, which meant that the houses of each district were numbered with consecutive digits. This same principle was used in setting up the numbering of the 1808 and 1853 systems [18].

The modern-day system in which house numbers are linked to streets via an odd and even principle was set up in 1876. In many cases, the order and numbering may have been subject to change. The contemporary building with the address ‘Prins Hendrikkade 73’, for instance, was known as ‘Oude Teertuinen 15’ in the year 1876. In 1853, within the footprint of the current building, four smaller houses were situated, each of which had its own house and district number (149, 150, 151 and 152) within the section M of Amsterdam.Footnote 5 We find their parcels neatly drawn in the 1832 cadastre, again each with its own specific cadastral address.

These historical addresses can be linked together by stating that the 1853 house number M 150 has the same location as the current Prins Hendrikkade 73 and vice versa, just as M 149, 151 and 152. We chose not to do this as this is conceptually not very accurate and lacks semantic refinement. As Tables 1 and 2 illustrate, the more historical mutations in parcels and house plots have occurred, the more fuzzy and convoluted the data becomes. Therefore, the older geographical infrastructure created by the HisGIS team in 2013 to provide the footprint of the 1832 cadastral parcels and buildings proved unsuitable and did not meet the requirements of the Amsterdam Time Machine project.

For the Amsterdam Time Machine GIS system, the HisGIS team introduced the principle of geographical coordinates as anchor points (represented by a formal identification-number, location-point or locatiepunt in Dutch) for all historical data with a spatial component that cannot be tied to specific geometries of buildings or plots. This system of location points had already been designed and tested as the key geometry in the so-called Time Machine for the Frisian cities, which was officially launched by the Fryske Akademy in May 2017, serving a pilot project concerning the town of Dokkum [42]. It was developed further within the Amsterdam Time Machine project between 2018 and early 2019. The concept of location points prevents common pitfalls of linking historical addresses with specific footprints as geometries of parcels or buildings as starting points. It makes it possible to gain insight into spatial continuity without the system being unyieldingly stuck to all historical mutations caused by joining, merging, demolition, splitting or aggregation of houses and their addresses.

Table 1. Linking historical addresses via the geometry of buildings or plots causes fuzzy and less accurate geographical relations between historical address identifiers. For example, the 1853 address M 151 relates to four 1832 parcel numbers, but it is unclear which one was the exact same plot.
Full size table
Table 2. Linking historical addresses via a point coordinate provides in more accurate one on one relations. Considering the Table 1 example, the 1853 address M 151 only relates to the correct and specific 1832 parcel plot.
Full size table

The HisGIS team created a dataset of over 52,000 location points (or coordinate sets) to form the basic GIS infrastructure for the Amsterdam Time Machine. These points are identified by (arbitrarily chosen) identification numbers which the addresses relate to.

The team chose to build this set of linked historical addresses from scratch and leave aside the already existing sets, such as the so-called renumber register of the Amsterdam city archive in which many historical Amsterdam addresses are linked over time. The reason for doing this is that, despite their comprehensiveness, these and other used systems are flawed by several of the common pitfalls outlined above resulting in an undesirable enlargement of the disarray of accurate spatial locations.

Considering the research focus area and period of the different use cases, the researchers involved decided to bring at least four historical house identifiers of the nineteenth and early twentieth centuries into this system. These are the cadastral parcel numbers of 1832, the so-called district numbers (wijknummers in Dutch) of 1853, the house numbers of 1876, and the potential later mutations and additions of the 1876 house numbers recorded in the year 1909. Per address, these four identifiers were linked by visually comparing geographically accurate georeferenced maps: the 1832 cadastral maps, the 1853 district maps, the so-called ‘Looman’ maps of 1876, and the 1909 Publieke Werken maps. In this way, a total of around 120,000 registrations through time are spatially elaborated by the 52,000 location points [40]. Recently, the system has been expanded with an additional set with the house numbers in use in 1943, which resulted in more than 53.000 new locations. We will shortly publish this set as Linked Open Data.Footnote 6

The location points and related house number set does not merely serve academic purposes. Thanks to its Linked Open Data (LOD) structure (see below), every researcher, professional, or hobbyist, may use the GIS infrastructure for his or her own purposes and research goals. Thus, the Amsterdam Time Machine GIS infrastructure can be used as a means of processing every dataset containing historical addresses of Amsterdam, to visualise data and analyse them spatially, in all kinds of fields and disciplines. More generally, this allows researchers to investigate the spatial correlation of certain phenomena, as in the use case discussed below, which focuses on the establishment of theaters and cinemas as cultural entertainment in relation to the socio-economic status of the inhabitants of the various Amsterdam neighbourhoods at the turn of the 20th century.

3.3 Linked Open Data Approach

To link the datasets in our project we use a technique called Linked Data [1]. Similarly to how websites can cross-reference each other, databases represented as Linked Data can contain cross-references to each other over the Web. For example, an address may contain many spelling variations, even within one and the same database. By representing an address via a URI (a unique identifier, possibly even approachable via the Web), that particular address can be referenced to without error and across multiple databases. For example, in the Adamlink projectFootnote 7, the URI https://www.adamlink.nl/geo/street/adriaan-van-swietenhof/64 refers to the street ‘Adriaan van Swietenhof’. Rather than having database entries such as ‘A. v. Swietenhof’ or ‘van Swietenhof’, we use the URI across various databases so we can link them. This approach is not much different from more traditional GIS databases where multiple entries for the same address are harmonised. The advantage of using URIs is that anyone can reuse them, even without our knowledge, since it is openly available as a public resource.Footnote 8 In doing so, datasets using the same URI are automatically linked, hence the term ‘Linked Data’.

The transformation of the data into Linked Data in this project facilitates future connections, among others with the images and other historical collections held at the various cultural heritage institutions.Footnote 9 As such, the Linked Data approach forms the foundation for the ‘deep mapping at scale’ that is our ambition.

As it is the central part of our database, we will now describe the data model used to create Linked Data for the HisGIS location points, while we describe the conversion to Linked Data for use case specific datasets in Sect. 4.2. To represent the temporal and geographical variant addresses as Linked Data, we applied the ontology design pattern by [25] for the space-time prism model [16]. Through its demonstrated application for CO\(_{2}\) measurements, we recognised similarities with our space and time variant addresses. In the CO\(_{2}\) case, cars drive around a city (space variant) and take multiple (time variant) measurements of the degree of CO\(_{2}\) in the air (value). To model this data, [25] suggest the use of so-called ‘control points’. A control point is somewhat like a ‘ping’ at which the car has a certain location, at a specific time and measures a value. The variant space, time and value measurements are attached to this ping and thus queryable. The latter is important to the space-time-prism model that in this way allows one to derive information on CO\(_{2}\) levels in places and at times where no measurements were taken.

In our study, the HisGIS location points are the control points or ‘pings’. The location points are by definition fixed in time, but observed at different points in time, and have different values for ‘address’. Figure 1 depicts for a particular location point how time variant addresses can be linked to these location points.

Fig. 1.
figure 1

Abridged visualisation of the HisGIS location points model based on the space-time-prism pattern

Full size image

The figure describes a tree-like pattern, with the location point as the trunk and the addresses as leaves. This means that any historical observation of an address, for example in an advert in a newspaper, can be linked to a location point. In addition, the model also allows us to see whether a change in address is actually related to a move or administrative change, a crucial difference in observation for life course research, for example. While there are but four time points covering the 1832–1909 period, through the space-time-prism model interpolations can be made for other years, as long as the values of the variable of interest are linked to the location points.

4 Use Case

4.1 Research Question

To test the Amsterdam geo-temporal infrastructure as a platform for urban historical research, we conducted a small pilot study on Amsterdam entertainment culture in the late 19th and early 20th century. The central question is to what extent the location of cinema venues, offering a new form of entertainment around the turn of the 20th century, can be understood in relation to existing forms of entertainment and to the social and economic composition of the neighbourhoods. Are cinemas established in the same neighbourhoods where theatres are located? What is the socio-economic status of those neighbourhoods, defined by income and by the dialects spoken there? What can such a comparison tell us about the status of cultural entertainment in late 19th- and early 20th-century Amsterdam? Such an interdisciplinary study builds on existing socio-economic historical analyses of the city and its inhabitants and enriches these with insights into cultural phenomena, such as language use and the appreciation of cultural entertainment.

Naturally, the location of theatres and cinemas in specific neighbourhoods cannot be seen as a direct indication of cultural appreciation, let alone of cultural consumption: these venues could very well be frequented by people from various parts of the city as well as outside of Amsterdam. At the same time, previous studies of cultural consumption in 19th-century Rotterdam and the Hague have shown that theatres and concert halls were mostly frequented by middle class and elite audiences, who used such visits as markers of social status [10, 13]. This indicates a certain correlation between the socio-economic status of a neighbourhood and the appreciation of the cultural venues located there.

Our analysis focuses on the shift in the Amsterdam entertainment landscape between two sample years: 1884, when the city had been expanded with a new, “19th-century belt”Footnote 10, and 1915, when the first permanent cinemas had been established in the city. For 1884, we study the correlation between the location of Amsterdam theatres and the socio-economic status of the neighbourhoods, marked by both the ‘elite density’ in those neighbourhoods and the dialects spoken there. For 1915, we studied, first, how the locations of the Amsterdam theatres have changed compared to 1884, and second, how these correlate with the locations of the cinemas and with the income of the inhabitants of the various neighbourhoods. We conclude with a tentative answer to the question what these patterns tell us about the status of cultural entertainment in late 19th- and early 20th-century Amsterdam.

4.2 Datasets

In addition to the geo-data described in Sect. 3.3, we have used various other datasets. The code we have used to clean these data and transpose them to Linked Data is available via Github.Footnote 11 For the time being the data can also directly be queried via the triple store Druid.Footnote 12 We will now describe these datasets in more detail.

Amsterdam Dialects. According to the 19th-century linguist Johan Winkler and the historian Jan ter Gouw a whopping number of nineteen dialects were spoken in 19th-century Amsterdam, distributed over various neighbourhoods [54], later plotted on a map [7]. Alongside these neighbourhood-specific dialects there were also sociolects spoken throughout the city, especially the slang of thieves and tramps. One of the questions we wanted to answer was whether the dialect variation in 19th-century Amsterdam was as great as Winkler and Ter Gouw claimed. For this, we attempted to reconstruct the dialect variation by collecting all Amsterdam language phenomena mentioned in 45 primary sources, including, for instance, Amsterdam word lists dated between 1800–1940. Subsequently, these language phenomena were manually entered into a database. The resulting database contains 8,020 entries categorised into the following language domains: words, names, idiomatic expressions, speech sounds, word formation, syntax, songs, and speech recordings. For each entry the information given in the original source was added, such as meaning, dating and bibliographical data. Finally, we tried to pinpoint each entry to a specific street or district, based on the information mentioned in the primary sources. In this we succeeded for 70% of the entries. However, it appeared that for fourteen of the nineteen dialects mentioned by Winkler and Ter Gouw, no concrete references to language phenomena could be found. As a next step, the RDF version of the data was modelled using the Lemon lexicon model for ontologies [6].

Occupational Structure, Tax and ‘Elite Density’. To derive information on people’s social and economic position in society, a key variable in sociological and history research, we reused data collected in the early 1980s and preserved thanks to the Dutch scientific data preservation organisation DANS [52, 53]. Historian Boudien de Vries took two samples (1854 and 1884) from the electoral roles of the city to obtain data on the Amsterdam elite. In addition to various other variables such as marital status and religion, she entered information on occupation, amount of tax paid and address of residence. We standardised the occupational titles onto the Historical International Classification of Occupations (HISCO) [28] through which we were directly able to attain information on the relative ‘position’ of these occupations in the occupational structure through HISCAM, an Historical CAMSIS scale [27]. Because it was also available as Linked Data, we could effortlessly add data on population size of every district in Amsterdam through the CEDAR project [36], allowing us to calculate a measure of ‘elite density’ by relating the number of elites to the population size per district. For the sample year 1915, we based the socio-economic status of the neighbourhoods on the 1915–1916 municipal tax records.Footnote 13

Theatre and Cinema. Data on the location and programming of theatres and cinemas are an important source for studying the broader social, cultural and economic contexts in which these forms of entertainment emerged and operated. Increasingly, such data are being digitised from secondary and primary sources, making them available for computational research [45]. The data for the Amsterdam theatre locations were collected by Charlotte Vrielink from the print publication [31] (Cf. [51]). The dataset is available as a CSV file at DANS.Footnote 14 The data on the cinema locations and their programming was collected from Cinema ContextFootnote 15, a historical data collection for the history of film culture in the Netherlands, which contains information on the cinemas active in Amsterdam and, to a large extent, the films screened there in the early 20th century [9]. For the purpose of this study, the geographic coordinates of the cinema venues contained in Cinema Context were replaced by more precise location points, some newly created by Vincent Baptist. The full dataset is available as a SQL data dump at DANS.Footnote 16

4.3 Method

The datasets used in our project, some originating from the early 1980s, were not readily available as Linked Data. We therefore used Python scripts and the CoW tool [38] to transpose our sources to Linked Data. CoW is a CSV-to-RDF converter aimed at researchers that have some computing skills, but are not developers themselves. Because basically anything can be represented via a URI, we were able to create Linked Data for the HisGIS location points, for the dataset on Amsterdam dialects (even including some sound samples of spoken dialect), the data on the locations of Amsterdam theatres and cinemas and the dataset on the occupational structure of the Amsterdam 19th-century elite.

To link the information from the separate databases, we uploaded the Linked Data in one central SPARQL Endpoint, in our case Druid.Footnote 17 SPARQL is a query language which allows one to retrieve a selection of data from one or multiple, linked datasets for further analysis. While SPARQL allows for some basic analysis and SPARQL query editors such as YASGUI [43] allow for basic (GIS) visualisations, because of the more complex requirements of our use case we mainly relied on offline tools with more functionality for the visualisation and analysis of our data. Thus, we created SPARQL queries to retrieve relevant subsets of the data that we subsequently analysed and visualised in R and QGIS. For example, we created a query providing information on the elite density, of which the result is downloadable as a CSV file, providing information on the number of elites, the population size and the geographical representation (polygon) per district. This CSV file was then imported into QGIS to visualise the elite density as a separate layer in our Deep Map. The visualisations of the various data layers in QGIS provided the basis for a qualitative analysis of the patterns observed.

4.4 Results

Figure 2 shows the 1884 map of Amsterdam, plotting the location of theatres against the demarcation of the areas of the city in which specific dialects were spoken and the ‘elite density’ of neighbourhoods.

Fig. 2.
figure 2

This map combines the three layers from the use case: dialect areas (approximate location indicated by dialect name), elite density (white: low; red: high density, classified using Jenks natural breaks) and the locations of theatres (cyan dots). Base map: Buurtatlas Loman, 1876, https://tiles.amsterdamtimemachine.nl/. (Color figure online)

Full size image

From this visualisation we observe that all theatres are located in areas with a relatively high elite density. We find no theatres in the relatively poorer areas of the cities, nor in the ‘golden bend’, the area of the canal ring which is the most affluent. One clear cluster can be observed in the Plantage district in the east of the city, a relatively new urban expansion for the well-to-do [29, p. 195-6], adjacent to the Jodenhoeks dialect on the Winkler dialect map. The area where ‘Kalverstraats’ was spoken also contains a number of theatres - this dialect can be seen as typically middle-class. One theatre is located at the Keizersgracht, where the upper class lived, speaking an upper class dialect. No theatres can be found in districts where typical working class dialects were spoken, with the exception of one theatre in the ‘Zeedijks’ or ‘Bierkaais’ area: De Vereeniging (located at Warmoesstraat 139). Therefore, we can conclude that the majority of theatres were located in the more affluent neighbourhoods and are conspicuously absent in the working-class neighbourhoods, if we regard both the elite density and the dialect information.

To investigate how cinemas found their place in this existing cultural entertainment landscape in the early 20th century, we first observe how the theatre locations have moved in the intervening 30 years. Figure 3 shows the shift of theatre locations between 1884 and 1915 from the inner city to the canal ring and the new expansion of Amsterdam South around Museum Square.

Fig. 3.
figure 3

Shift in the location of Amsterdam theatre venues between those active in 1884 (cyan dots) and those active in 1915 (dark blue dots for theatres that were still active since 1884 and green dots for theatres that had opened since 1885). Base map: Dienst der Publieke Werken, 1909, https://tiles.amsterdamtimemachine.nl/. (Color figure online)

Full size image

Plotting the 1915 theatre locations against the economic status of the neighbourhood (Fig. 4), we observe that some theatres remain in the locations of 1884 (in particular in the area where ‘Kalverstraats’ is spoken and in the Plantage district), but in general, the theatres have moved in tandem with the city’s expansion. Contrary to 1884, we now see two theatres appear in the popular ‘Jordaan’ district, but otherwise the pattern is the same: most theatres are still located in the more upmarket neighbourhoods - i.e. neighbourhoods in which a relatively high percentage of inhabitants pays the highest tax tariff. As in 1884, the most expensive areas of the city do not contain theatres.

Fig. 4.
figure 4

Amsterdam theatre locations in 1915 (dark blue dots) plotted against the economic status of the neighbourhood. The white areas are neighbourhoods where none or only a very small percentage of inhabitants were in the highest tax bracket (over 5,100 Dutch guilders); the redder the neighbourhood, the higher the tax percentage (classified using Jenks natural breaks). Base map: Dienst der Publieke Werken, 1909, https://tiles.amsterdamtimemachine.nl/. (Color figure online)

Full size image

When looking at the location of cinema theatres in 1915, plotted against the economic status of the neighbourhoods in Fig. 5, compared to theatres we clearly see much more cinemas in the areas with no or few inhabitants in the highest tax bracket areas, here used as a proxy for indicating the areas with poorer inhabitants.

Fig. 5.
figure 5

Amsterdam cinema locations in 1915 (magenta dots) plotted against the economic status of the neighbourhood. The white areas are neighbourhoods where none or only a very small percentage of inhabitants were in the highest tax bracket (over 5,100 Dutch guilders); the redder the neighbourhood, the higher the tax percentage (classified using Jenks natural breaks). Base map: Dienst der Publieke Werken, 1909, https://tiles.amsterdamtimemachine.nl/. (Color figure online)

Full size image

Comparing the 1915 theatre and cinema locations, as plotted against the socio-economic status of the neighbourhoods in Fig. 6, we see striking differences in their location patterns: the cinemas are located in ‘older’ areas of the city, for example in the North-East corner area of the city where in the 19th century people spoke ‘Nieuwendijks’, ‘Haarlemmerdijks’ and ‘Jordaans’, all of which have been identified as popular dialects. And in the relatively up-market Plantage district, no cinemas can be found, whereas in the Jewish quarter, where in 1884 the ‘Jodenhoeks’ popular dialect was spoken, we find no fewer than three cinemas (including the ‘Tip Top theater’ that was known as a Jewish cinema) and no theatres.

Fig. 6.
figure 6

Amsterdam theatre (dark blue dots) and cinema (magenta dots) locations in 1915 plotted against the economic status of the neighbourhood. The white areas are neighbourhoods where none or only a very small percentage of inhabitants were in the highest tax bracket (over 5,100 Dutch guilders); the redder the neighbourhood, the higher the tax percentage (classified using Jenks natural breaks). Base map: Dienst der Publieke Werken, 1909, https://tiles.amsterdamtimemachine.nl/. (Color figure online)

Full size image

In conclusion, we can note that there is little overlap between the location of theatres and cinemas: the new form of entertainment did not connect to existing venues to create entertainment districts but found a place in other, less affluent areas of the city. This may be explained from the fact that most of the 1915 theatres had already been established in the period 1880–1890, but can also indicate that the new form of entertainment was seen as less respectable and geared towards the lower income groups.

At the same time, the opposition between theatres and cinemas should not be overstated. Within both sectors, there were venues that catered for higher and for lower segments of the markets. For example, there were ‘chic’ cinema theatres in the ‘Kalverstraats’ dialect area that catered for a middle class audience by offering the latest premieres (perhaps not accidentally the one area which contains both cinemas and theaters), and cheap theatres that explicitly catered for a working-class audience (such as the Rozentheater in the Jordaan district).

5 Discussion

Our effort to create a Semantic Deep Map appears to have been a successful undertaking. We were able to combine multi-typed datasets, from various disciplines and answer a question with the help of these sources.

However, a first point for discussion is that we were unable to follow the Linked Data pipeline up to the point of the analysis because we had to switch from the online dataset collection and dataset querying to the offline data visualisation in QGIS. All the steps needed to get this visualisation, are not stored and therefore not reproducible. In part, this might be improved upon by more advanced SPARQL skills, but for another part it might also be a limitation of the query language, which is not optimised for geographical queries. A solution for this may lay in the use of Jupyter Notebooks, in which one can apply multiple languages (R, Python, SPARQL) to create, query and analyse Linked Data. With a Jupyter Notebook, each language could then be used at its core strength. In addition, all code in the chain could be run for a complete reproduction of the research.

Whilst the Amsterdam Time Machine combined dataset sheds new light on the late 19th- and early 20th-century entertainment industry, it by no means can provide all answers. A core limitation is that the data is incomplete and more a series of ‘snapshots’ than a continuous record. Thankfully, more digitised data is becoming available every day, which can be linked to the current Amsterdam Time Machine dataset.

Furthermore, whilst the data does provide some information on trends, it does not allow us to switch seamlessly between micro, meso and macro levels. To do this, more information is needed, for example on particular households and their staff living at an address and information on what places besides their residence these people frequented. One limitation of the linguistic data is that rich people living in the most expensive neighbourhoods employed staff that may or may not have lived in their house and spoke another dialect or sociolect than their employers.

When we connect for example the socio-economic data to the dialect data, or the cinema and theatre locations to socio-economic data, we do not know if the people who lived in those areas were actually the ones visiting those venues. While our data does indicate where cinemas and theatres were located, and that in combination with the social status information about the neighbourhood we see correlations between ‘higher’ and ‘lower’ cultural venues, our data cannot tell us who actually visited these venues. Play bills, ticket sales information (as cinemas and theatres both used to sell price differentiated tickets to different seating areas), but also mentions of shows in letters and diaries could provide additional insights here. There are several projects ongoing that aim to make this type of data digitally available.Footnote 18 Our efforts in this project can serve as a blueprint and anchor for adding such data layers.

6 Conclusion

With space as a connecting factor, the Amsterdam Time Machine provides a concrete illustration of the research potential of linking social and economic data with cultural data, allowing researchers to study specific historical and cultural phenomena against the background of broader societal developments. Our approach, based on the identification of location points and the transformation of data into Linked Open Data, allows for the building of a rich tapestry of information about a city and its development over time - both physically and in the way it has been used and experienced. As such, it points towards a ‘deep mapping at scale’: the creation of a geographical-temporal infrastructure that allows for the querying of multiple types of data in a scalable manner, navigating between the broader patterns of urban life and the micro level of individual people, places and events.

We conclude that in its present form, the platform provides heuristic support for interdisciplinary digital humanities research, in that the combined visualisation and exploration of different types of data exposes correlations that lead to new hypotheses about historical urban entertainment culture. Our pilot use case showed that in early 20th-century Amsterdam there is little overlap between the locations of theatres and cinemas and that, where theatres were located in relatively affluent areas of the city, cinemas were located in less affluent districts. This generated the hypothesis that cinema as a new form of entertainment was seen as less respectable than theatre and hence ‘took place’ in less affluent areas of the city. As such, the use case demonstrates the potential of deep mapping as a method that invites experimentation to further test and refine hypotheses.

We are currently in the process of extending the temporal scope of the location points forward in time, to 1943, and back in time to 1647. In the near future, the historical addresses as a dataset and fundamental stepping stone for the Amsterdam Time Machine and its infrastructure will be semantically embedded within the scope of ontologies for historical geographical data on the one hand, and the domain-central HisGIS ontology of the historical cadastral entities on the other. In parallel, we are setting up a sustainable environment for the infrastructure, on the one hand by the creation of sustainable endpoints and a common infrastructure, on the other by establishing an editorial office and editorial board for the Amsterdam Time Machine geodata to ensure that a common strategy and procedure is followed. We encourage future research projects to build upon the Amsterdam Time Machine infrastructure: each new layer of information uncovers another part of the rich history of Amsterdam and its citizens and provides scholars with a new lens through which to study our past.