Practices of discursive urban place-making in Brooklyn, New York: (hidden) digital and embodied discourse

2.1 The model of urbanity

A “spatial dimension” (Busse 2019: 20; Busse and Warnke 2015) may refer to the geographical coordinates of the physical environments. If human beings interact in a physical space, a space becomes a place endowed with meaning (Busse 2019; Cresswell 2015; Warnke 2017), which is constantly negotiated and contested. As the city, like any other space, is linguistically loaded, discourse in an urban space, that is, the use of linguistic and other semiotic modes in a context of urbanity, is one such social process of place-making. Urban spaces are parameters of linguistic variation, and language is “also a parameter for the construal of urbanity by means of place-making practices” (Busse 2019: 21; see also Warnke 2017).

In Busse and Warnke’s (2015) model, the adjective urban describes a set of values of a place, neighborhood, or city. Conceptually, it denotes an evaluative rather than a descriptive relation. Hence, the terms urban/urbanity are prone to act as dynamic meaning-making concepts (see Busse and Warnke forthcoming) in complex and dense spatial constellation in a city. Following Lefebvre (1974), three modes – 1) dimension, 2) action and 3) representation – constitute the urbanity model (Busse and Warnke 2015) shown in Table 1.

If we transfer the conceptual configurations of the urbanity model to the photograph displayed in Figure 1, we see two discourse contexts of this photograph, one is who and what are in it and what they are doing; the other is that this is a picture made by a photographer who has intentionally included the photograph for his public exhibition. In the urbanity model, dimension describes spatial configurations. In the photograph, the dimensional configurations are materialized through Jerell Willis’s body and the space in which the photograph with his son was taken, namely, on the Brooklyn Bridge, which connects the borough of Brooklyn with that of Manhattan.

For the second discourse context – that of the photographer choosing this photograph for his exhibition – the dimensional spatial configuration is that of depicting black fatherhood a) in New York and b) in a material exhibition. Action carries interpersonal dimensions and describes what agents do in space. In the photograph, Jerell Willis walks the Brooklyn Bridge carrying his son and displays his naked body, with the peculiar Brooklyn tattoo as semiotic place-making on the body. As a photographer, Zun Lee portrays an intimate moment between the father and his son. He also gives voice to a former Brooklynite of color. Representation describes the ways in which semiotic practices – linguistic or other – represent and construe meaning and places. In the photograph, this place-making is construed through multivariate sign-making – Brooklyn in the artistic font on the back of a human torso and by walking the bridge – and the specific and foregrounded design of the tattoo. Through the choice of the medium of a photograph on the discourse dimension of the photographer the meaning-making practice becomes not only permanent but also public.

In the urbanity model, the three modes are further characterized by six distinct features: size, density, heterogeneity, simultaneity, multiformality, and multi-semiosis. Size and density further modify the dimensional mode because they allow for a more detailed description of the scope of an urban space under investigation. To refer this to the discourse contexts of the photograph, the features of size and density are metaphorically evoked through the Brooklyn Bridge – the cultural icon of the ethnically most diverse and densely populated megacity New York. Heterogeneity and simultaneity account for the fact that any action may be performed simultaneously and by various agents, as can be seen in the photograph for both discourse contexts. Despite the fact that the borough of Brooklyn is still part of Jerell Willis’s identity, as a person of color he was pushed out of Brooklyn. Multiformality and multi-semiosis ensure that any semiotic practice or any incident are realized and construed by a multitude of forms which consist of different modes – tattoos, walking the bridge, the choice of creating a black-and-white photograph. The denser, more heterogeneous, and inter-semiotic a city is, the more urban it will be, and the more linguistic and semiotic variation will be encountered.

The photograph with Jerell Willis and his son Fidel illustrates, on various levels, the complexity of discursive urban place-making: fugitivity, because he was forced to leave his neighborhood, and futility, with the next generation on his arm, and, at the same time, maintaining black cultural belonging to a borough that no longer affirms black people’s place in it, or at least not in those gentrified Brooklyn neighborhoods which are directly facing Manhattan. At the same time, it also negotiates precarious intimacy and relationships between him as the father and his son in a very public and even exhibited way. Added to this is that Brooklyn and the Brooklyn Bridge have various connotations for different people from different ethnic backgrounds.

Urban space is also characterized by the declarative function in Searle’s (1976) speech act model. Linguistic and semiotic practices in urban spaces may be endowed with a declarative illocutionary force which has the potential for changing – in the sense of a declarative city – urban reality from space to place (Cresswell 2015) because they are embedded in the modes of dimension, action, and representation. These urban semiotic place-making practices and semiotic interventions may become indexical and have enregistering potential in the sense that a community of practice – temporarily or not – has endowed linguistic practices with certain social values (Agha 2003, 2005; see also Johnstone 2009, 2017; Monka et al. 2020). In other words, all linguistic and semiotic acts – not just dialect features alone (Johnstone 2009) – may be associated or linked with particular social values and may be used to do social work (Busse 2019). Jerell Willis’s tattoo is a performative act in which Brooklyn is created as an artsy place and in which, through materially engraving art on his naked torso, the declaration also creates Brooklyn. The photographer’s capturing of both his body and the moment of walking the bridge are equally declarative, as these processes lend this act permanence and publicity (see also Blommaert 2013; Britain 2018; Eckert 2012; Warnke 2017).

2.2 Discursive urban place-making and foregrounding

The question arises when we know that social value is meaningfully indexed with certain linguistic and semiotic patterns in various contexts of urbanity. For this, it is useful to unlock the interplay between quantification as seen in statistical foregroundedness, for example, as well as in the qualitative investigation of singular, multi-semiotic discursive practices existent in urban spaces. In his book Language in literature: Style and foregrounding, Geoffrey Leech regards the “varied applications of the concept of foregrounding” (2008: 4) as key to analyzing and explicating literary texts. On the basis of his definition of style as language variation according to varying circumstances, Leech (2008) illustrates the two dimensions of foregrounding, deviation and parallelism (Hakemulder 2020). For example, foregrounding may be “deviation, or departure, from what is expected in the linguistic code expressed through language […]” and that “functionally, it is a special effect or significance conveyed by that departure” (Leech 2008: 3). In other words, conventions of language usage form the background for how linguistic structures become prominent. Hence, assessing foregrounding – through parallelism and deviation – is always also a comparative exercise. If style is variation in language usage, then each linguistic/semiotic choice is stylistic, multifunctional, and therefore socially meaningful in context and within (or devoid of) certain intra – and intertextual norms set up in space, time, or by certain communities of practice.

In order to further assess the extent of foregrounding mechanisms that can be discerned in the Jerell Willis example, I draw on my earlier work (Busse 2019). I was able to outline the tentative patterns of discursive urban place-making in a small corpus of 33,117 words consisting of semi-structured interviews and conducted in various gentrified neighborhoods of Brooklyn (such as Brooklyn Heights, Williamsburg, or Park Slope) in 2012. With the help of a keyword analysis, which assesses a text’s aboutness through a statistical comparison of lexical items in the text under investigation with a reference corpus, I show that Brooklyn as a proper name is statistically marked in my interview data and I claim that it is used generically to enregister (Johnstone 2009) Brooklyn© as a brand. In the interview data, Brooklyn or the respective neighborhoods are constantly compared with Manhattan as a borough and the Brooklyn© values created are coolness, authenticity, a community – and neighborhood feeling, or striving for sustainability. Here is one example from a Brooklynite who answers the interview question “How would you describe this place?” by various generic references to Brooklyn as a brand:

(1) Like a lot of like it’s very like homegrown, like grassroots, like a lot of people are like you know, if it’s like you know oh this isn’t made in Brooklyn or handmade in Brooklyn, it’s like a cool […] a little bit of an authentic city (interview with female Brooklynite, age group 30–40).

There are striking discursive similarities between the place-making practices/features identified in the interview corpus and in the photograph. The reference to Brooklyn on Jerell Willis’s back also construes Brooklyn© as a brand, with the variation that it is on the body and multimodally and artistically crafted with the depiction of the iconic Brooklyn Bridge. However, the embodied art on Jerell Willis’s back might also stand for a reclaiming of space following severe processes of gentrification already taking place in Brooklyn in 2012, when the photograph was taken. Brooklyn was then and still is also one of the most ethnically heterogenous neighborhoods of New York (Fessenden and Roberts 2011). While Brooklyn is described by some as an up-and-coming or “authentic” (Busse 2019) borough in which neighborhoods like DUMBO change their character from “artsy to digital DUMBO” and where many warehouses are turned into expensive condominiums, not much is said of a) those inhabitants who are pushed out of gentrified places or b) different ethnic communities to whom Brooklyn is home, too. Perhaps what we see here is a foregrounded (re-)claiming of place, even a resistance against the backdrop of “urban” place-making – which has so far been either romanticized or neglected and led to massive changes in Brooklyn – some call it gentrification and renewal, others call it dispossession and displacement.

Foregrounding through parallelism is therefore construed by the generic multimodal reference to the pattern of Brooklyn as a name on all dimensions of the urbanity model. This comparative exercise is crucial to capture how the semiotic place-making on the body of Jerell Willis’s back relates to other, semiotically and diachronically distinct discursive urban place-making practices collected before and after the photograph was taken. The comparative exercise is also crucial to Busse and Warnke’s (2015) urbanity model, both methodologically and theoretically. For the former, it highlights the mixed method approach – uniquely combining disciplines such as corpus linguistics, sociolinguistics, ethnography, urban geography etc. – and the claim to include in investigations of urban discursive place-making, for example, both quantitative and qualitative data and a diachronic perspective. For the latter, that is, theoretically, I rely on the Leechean notion that patterns of foregrounding can only be discerned through comparison of various linguistic practices in context.

Following the outline of the theoretical model of urbanity and its operationalization through the analysis of the photograph in Figure 1, the next session will introduce the heiURBAN corpus and data investigated further for this study.

4.1 Wi-Fi SSIDs in Brooklyn, New York

In this section, I will present both the methodology for collecting Wi-Fi Service Set Identifiers (SSIDs) for the heiURBAN corpus and show how that compilation process has been further developed. I will then present some of the first research findings of the analytical procedures which evolved around a) ways of identifying linguistic structures, b) categorizing meaningful Wi-Fi SSIDs and c) describing their functions as acts of discursive urban-place-making. Suggestions for categorizations of Wi-Fi SSIDs are based on the pioneering work Ingo Warnke and I did in Berlin in 2012 (see Busse and Warnke 2016).

Wi-Fi SSIDs are the names people give to their wireless networks. With that, I do not mean the noisy data, primarily consisting of unchanged default network names, illustrated in Figure 4 and which I also found during our 2017-field trip.

Figure 4:

Noisy data in the Wi-Fi SSID collection process.

Contrary to these, I refer to the semantically meaningful labelling such as: “Don’t throw food in our yard”, found in Bedford Stuyvesant, “BackInBrooklyn”, recorded in Williamsburg/Broadway Triangle, “Dumbo Time Capsule” recorded in Dumbo, or “Fuck Donald Trump”, recorded in Williamsburg.

Applying the urbanity model (Busse and Warnke 2015) to these examples, their dimension is the location in which these Wi-Fi SSIDs were found. Instead of just recording the neighborhood and their coordinates, my team and I recorded geocoordinates each time a network was encountered. As regards the dimension of mode, it is obvious that human beings have decided to exploit that semiotic potential given in Wi-Fi SSIDs and decided to label them. Their representation is primarily linguistic and human beings’ linguistic creativity for naming is beyond limits and perhaps reinforced in an initially invisible sphere, as these can only be seen using a Wi-Fi-enabled device such as a smartphone (Stelthove 2019). They may also be changed continuously. Therefore, they are mutable and ephemeral. It is now possible to sample, compile, and code Wi-Fi SSIDs because wireless networks following the 802.11 standard (commonly known as Wi-Fi or WLAN) are identifiable by a user-defined name (IEEE 1999: 55). Interestingly, the standard does not specify the encoding of the SSID.

While this type of linguistic data is extremely interesting for research in urban linguistics, collecting the data is first of all a momentous and certainly not a trivial task. The methods of collecting and compiling Wi-Fi SSID data were both used and refined following a field trip to Brooklyn in 2017. Conducting a pilot study, my team and I used our smartphones to document SSIDs in certain areas by manually copying them out of the network connection manager. Geographical coordinates were then, whenever possible, automatically approximated.

My team and I then devised a new approach. In the field of information security, wardriving is a common practice involving the “the act of moving around a specific area, mapping the population of wireless access points for statistical purposes” (Hurley et al. 2007: 3) in order to identify vulnerable or weak networks. It has been repurposed for other types of research: semantic information for marketing purposes (Seneviratne et al. 2015), to investigate Wi-Fi use in public settings such as parks (Forlano 2008), as well as for ethnographic studies on social relationships (Hampton and Gupta 2008). However, this technique has seldomly been fused into a corpus or used by (socio)linguists to tap into hidden language data and urban place.

Therefore, for our 2017-field trip, my team and I automatized the SSID collection process by using WiGLE Wi-Fi Wardriving (Carra et al. 2020), an open-source mobile application for scanning and mapping wireless networks. In order to make the collection process as seamless as possible, all researchers carried a smartphone or tablet running the application which was constantly gathering all available access points including metadata such as the current location and the current signal strength. Therefore, the researchers were able to record all this data while walking the streets, interviewing people, and taking photographs of the linguistic landscape in those assigned streets representative of the specific neighborhood. While this approach worked very well, the researchers still needed to remember to keep their devices close to buildings due to the limited signal coverage of only 25–100 m (Cooklev 2004; Harwood 2011). At the end of each day, the data from all devices was transferred to an external storage device. Following the fieldwork, the individual datasets (multiple files per day per researcher) were combined, post-processed, and ultimately added to a SQL database. Since SSIDs are not unique, the unique Basic Service Set Identifier of the access points to disambiguate networks was used. Given that there are enough data points on one particular network, in many cases my team and I were also able to triangulate the position of the access point based on signal strength, giving us more detailed information about the origin of the message.

While many of the fundamental technical challenges can be solved this way, there are still a number of open questions. Firstly, there is no interaction with the creators of the SSIDs, and no socio-demographic information beyond an approximate location can be obtained. Secondly, SSIDs can be extremely ephemeral. Therefore, every SSID dataset has to be treated as a very specific kind of sample, which is highly dependent on the timing and the movement of the researchers. Lastly, one could debate the (non-)public nature of these names from a privacy perspective and whether it is ethically responsible to capture, locate, and analyze them.

4.2 Networks of Tweets with #Brooklyn node

4.2.1 Data compilation

Tweets like those displayed in (2) and (3) below represent complex multimodal ways of creating urban places, because #-discourse establishes meaningful #-connections between one or many hashtags and themes. In example (2), the Brooklyn neighborhood Bushwick is positively identified with street art as “the best neighborhoods for #streetart”. In the second example, Brooklyn as a borough in general and Bushwick as one neighborhood in particular are criticized as “unethical” in relation to the real estate business.

(2) “I may very well live in one of the best neighborhoods for #streetart ❤ So lucky. #bushwick” (Fordyce 2017).

(3) “So unethical #bushwick #brooklyn #realestate #residentialrealestate

https://t.co/gJZxx0lyBY” (@100Bogart).

We have already encountered both of these contradictory patterns of urban discursive place-making: a) praise of Brooklyn (neighborhoods) for their artsy scene and b) severe criticism of gentrification processes.

Referring these practices to the urbanity model, we can assess that the dimensional mode is realized by when and from where the Tweet was posted. Action and representation strongly interact despite the restrictions on the number of signs used for Tweets. Unlike with Wi-Fi SSIDs, where the sender cannot usually be identified, the actional mode can be related to the persona posting the tweet. Similar to the Wi-Fi SSIDs, the representation of Tweets can be realized multimodally and in complex meaning-making networks through re-posting, liking, re-tweeting, hashtagging and (re-)connecting. It is due to this complex weaving of Tweets that the features of the modes of the urbanity received additional complexity.

For the purpose of this paper, I shall analyze the individual #Brooklyn Tweets and interpret their function or contextualize them. On the basis of big data analysis, the aim is rather to try to assess quantitatively the rhizome-like semantic #Brooklyn- interconnection to find out if and how these form digital patterns of discursive urban place-making. The method used is network analysis.

Over the course of over three years, starting in January 2017, I have compiled and stored Brooklyn and Brooklyn-related Tweets and hashtags in order to investigate how these are tagged further and related to other topic markers (Zappavigna 2015). Currently, there are approximately 120 million Tweets and metadata. Although, this is still in the realm of small Big Data, it is a challenge to analyze or even view the data with traditional (corpus linguistic) software because of the dynamic nature and the highly intricate metadata situation. Another challenge lies in cleaning the data by filtering out duplicates, unrelated, and nonsensical Tweets from the dataset.

The sampling process, which led to the samples underlying the visualizations in Figures 5 and 6, is relatively complex, too, and possibly introduces certain biases. Generally speaking, I consider the population to be all Tweets in some way referring to Brooklyn that are being tweeted. For this particular project, my team and I defined the population more restrictively to be all Tweets referring to a) Brooklyn and b) being posted in 2019. Due to the fact that it is impossible to collect all Tweets being tweeted, the heiURBAN Twitter corpus is already only a small sample of the population.

Figure 5:

Co-occurrence network (50,000 Tweets from 2019, all containing the hashtag #Brooklyn).

Figure 6:

Co-occurrence network (100,000 randomly sampled Tweets from 2019; #Brooklyn and directly connected nodes are blue).

In order to compile the heiURBAN Twitter corpus, public Tweets, based on a predefined set of search terms and hashtags as well as on geolocation,^[3] have been continuously collected using Twitter’s standards streaming/filtering Aplication Programming Interface (API)^[4] (using TweetPinna to access the API and store the Tweets continuously). It is important to note that this API (in contrast to the commercial PowerTrack API with access to the full Twitter firehose) has various limitations and has been changed multiple times since 2017.

Unfortunately, collecting Tweets this way does not ensure that all collected Tweets are actually from the population. Also, the smaller the set of search terms and the more restrictive the filters are, the higher the chances of missing Tweets belonging to the population. Therefore, in order to capture more Tweets and to increase our chances of actually sampling the population, hashtags and search terms referring to New York City (and NYC) were added. While this approach adds a large amount of noise to the dataset, it allows us to carefully filter Tweets after the collection process at our own pace although Zappavigna (2018) points out that sampling based on hashtags always is impacted by the self-selecting nature of users using hashtags. Unfortunately, due to changes in Twitter’s API as well as server maintenance, there are some days during which no data has been collected.

For 2019, the period in question, the database contains 339,437,44 Tweets. However, as indicated above, given the wide search parameters, many of these sampled Tweets are not part of the above-mentioned population. Usually, it is now necessary to apply sophisticated filter and categorization approaches in order to select those Tweets, which fit the population definition. For the visualizations above, we created one filtered and one unfiltered sample to demonstrate this. However, since links to #Brooklyn are in focus, non-related Tweets will not impact the relationships in question, but only frequencies.

Out of these 339,437,44 Tweets, Figures 5 and 6 show two random samples. The sample for Figure 5 contains 50,000 Tweets, each of them containing the hashtag-type #brooklyn (case-insensitive and sensitive to this spelling, so not “#Bklyn” or “#Bkln”.). While this choice skews the frequency of the #brooklyn hashtag and limits the sample heavily, it will definitely ensure that the sample only contains Tweets from the actual population (Figure 5). The second, larger sample (Figure 6) contains 100,000 Tweets and has no restriction on content. Thus, this second sample is more representative of the whole corpus but will most likely include Tweets which are not part of the population as well as Tweets without any hashtags.

The visualizations show hashtag co-occurrence graphs (networks). The graphs have been constructed by looking at which hashtags are co-occurring in the sampled Tweets. Technically, this has been achieved by using a custom-built Python tool based on NetworkX (Hagberg et al. 2020). In the graphs, each node represents one hashtag (all hashtags are lowercased), and edges represent if and how often hashtags co-occur within Tweets. The size of the nodes is a scaled representation of the hashtag frequency in the sample. Similarly, the width of the edges is a scaled representation of the frequency of co-occurrence.