A Technique to Infer Symbolic and Socio-Symbolic Micro
Artem Antonyuk1,2*, Kseniia Puzyreva1,2 [0000-0001-86 99-9553], Darkhan Medeuov1, and Ni-
kita Basov1,2 [0000 -0003-3630-6119]
1 Centre for German and European Studies, St. Petersburg University, St. Petersburg, Russia
2 Faculty of Sociology, St. Petersburg University, St. Petersburg, Russia
Abstract. The interplay between symbolic and social structures in groups is often
analysed at the whole-network level of their semantic and socio-semantic net-
works, e.g. via comparison of graph distributions, multidimensional scaling, or
QAP correlations. Meanwhile, the interplay between the symbolic and the social
operates through the usage of signs (e.g. words) and their associations by inter-
acting individuals. Hence, structural properties of the whole network can be ex-
plained by analysing specific instances of symbolic and socio-symbolic micro
patterns – elementary configurations linking signs, and signs and individuals –
occurring in practical contexts. This paper introduces a technique and a customis-
able pattern retriever tool (an R script) to (1) programme socio-symbolic patterns
of theoretical importance, (2) use them as ‘search terms’ to query network data,
(3) extract from the data instances of the patterns and text quotes corresponding
to them, (4) store and represent these instances and quotes in a form convenient
for their subsequent qualitative analysis – to uncover the contextual meanings of
the patterns. We illustrate the proposed technique with an analysis of a mixed
dataset on the interplay between expert and local symbolic structures in the con-
text of social structures of two local groups engaged in flood risk management in
Keywords: Pattern retrieval, Socio-semantic network, Symbolic structure, So-
Increasing attention is being paid to the co-evolution of symbolic and social structures
[1–6]. The interplay between these structures in social groups is often approached at
the whole-network level of structure and content of groups’ (socio-)semantic networks,
for instance, via comparison of graph distributions [7, 8], multidimensional scaling and
QAP correlations , or hyperbolic spaces . Meanwhile, the symbolic and the so-
cial interplay through the exchanges of signs (e.g. words) and their meaningful
associations in utterances by particular individuals interacting in practical contexts [11,
12]. Therefore, structural properties and content of symbolic and social structures
should be explained through an analysis of their micro patterns – elementary configu-
rations linking signs, and signs and socially tied individuals – as they occur in concrete
verbal expressions and interactions against the backgrounds of broader cultural and so-
cial contexts [13, 14]. So far, however, there has been a lack of techniques and tools to
facilitate the systematic extraction of such patterns from empirical data.
The present paper addresses this gap and introduces a technique and a customisable
pattern retriever tool – an R  script – for semi-automatic computer-aided extraction
of (socio-)symbolic micro patterns from empirical network data. The technique and the
tool enable the specification of patterns of theoretical importance, accurate and fast
retrieval of instances of specified patterns from semantic and socio-semantic networks,
and extraction of textual contexts (phrases and sentences) of the retrieved patterns’
components from the data for subsequent qualitative analysis [see 14]. As the majority
of operations are carried out within the R environment, this tool minimises the possi-
bility of unintentional errors and data loss because of conversion between different data
formats. The technique and the tool can retrieve patterns from longitudinal as well as
The paper is organised as follows. First, we conceptually introduce patterns of the
interplay between symbolic structures, and between symbolic and social structures.
Second, we introduce the technique and the tool for extracting instances of patterns
from the data. Finally, we illustrate the technique and the tool using our mixed dataset
on local and expert groups engaged in flood risk management gathered in England in
2 Symbolic and Socio-symbolic Patterns
Social groups use signs to refer to objects, actors, actions, and situations. By associating
signs in particular ways, groups express their specific identities and perspectives on
reality . For example, by associating ‘river’ with ‘flooding’, group members ex-
press their shared understanding of a river as something that might flood surrounding
areas. This meaning of the river may be irrelevant to groups that occupy riverbanks but
never experienced floods. Signs and associations continuously used by a group consti-
tute a symbolic structure of this group. The signs and associations between them are
the focal subject for the analysis of the interplay between symbolic structures of differ-
ent groups .
The interplay between symbolic structures involves the mutually induced reproduc-
tion and/or change of signs and their associations used by different groups. For instance,
consider symbolic structures of expert groups, e.g. scientists, politicians, or represent-
atives of NGOs. Usually developed in a comprehensive and professional manner, sym-
bolic structures of expert groups have the authority to identify issues and propose solu-
tions in the corresponding fields of expertise . These definitions, often expressed in
scientific research, political programmes, and laws, are imposed through authoritative
language to be adopted and enacted by local groups (e.g. ordinary citizens, indigenous
people) in a field [e.g. 18].
Meanwhile, symbolic structures of local groups are developed in a more spontaneous
manner in everyday practice [e.g. 19, 20]. These symbolic structures include local news
and rumours, stories, jokes, etc., which are mostly relevant for a particular local group
and have little relevance for others. Owing to the difference in social statuses of expert
and local groups, locals are subjected to significant institutional pressures to adopt cer-
tain meanings (such as definitions of situations, issues, models, and best practices.)
from experts . Thus, local symbolic structures can be regarded as ‘dependent’ in
relation to ‘independent’ expert symbolic structures.
At the same time, local groups often resist institutional pressures [22–24] and rein-
terpret elements of expert symbolic structures, instantiating them locally [14, 25]. This
way, locals simultaneously meet institutional expectations [26, 27] and preserve their
own definitions of situations and of their place in them [22, 26]. Transformation of
‘dependent’ symbolic structures under the influence of ‘independent’ symbolic struc-
tures reveals itself in language and can be traced using symbolic patterns that capture
addition, change, and removal of signs and meaningful associations between mutually
defining signs over time.
Individuals constitute their common perspectives on reality by using and combining
signs as they interact in a group and refer to its common context . They may repro-
duce the existing signs and associations between signs, recombine signs, or introduce
new ones [28, 29]. This mostly happens in the context of dyadic and triadic social ties
between group members and, therefore, the group’s symbolic structure relies on the
structure of social ties within the group [30–34]. Hence, the effect of social structure
on the symbolic structure should be controlled for when examining the interplay be-
tween symbolic structures. It can be traced through socio-symbolic patterns that com-
bine social ties between individuals with signs and associations between signs they use.
Based on the existing literature, we theorised a number of symbolic  and socio-
symbolic  patterns of the interplay between different groups’ symbolic structures
in the context of their local social structures. For illustrative purposes, the further
presentation of our pattern inference technique relies on one pattern of each type.
The symbolic pattern loose coupling reflects how a sign from one symbolic structure
is reinterpreted in another symbolic structure . This pattern represents a process
when actors from one group (e.g. locals) reproduce and reinterpret signs used by an-
other group (e.g. experts), such as specific terms, in the context of their own symbolic
structure, fitting them to their purposes. More specifically, it implies that a group’s sign
used at t1 is reproduced by another group at t2 in a new association with a pre-existing
sign specific to the second group (see Fig. 1).
Fig. 1. Symbolic pattern ‘loose coupling’. Blue diamonds = signs used only by locals; red dia-
monds = locally reproduced expert sign; orange diamonds = sign used by experts and locals;
red line = expert-specific sign association; blue line = new local-specific sign association.
As reflected in Fig. 1, locals may appropriate from experts the idea of producing plans,
start to use the sign ‘plan’, and adapt it to their own local context by associating it with
the sign ‘neighbourhood’. Simultaneously, they ignore the original experts’ understand-
ing of this idea represented by the association between the signs ‘plan’ and ‘manage-
The socio-symbolic pattern contagion implies the reproduction of signs and/or asso-
ciations between signs used by a single member of a group by other group members
who were not yet using them. Such signs and/or associations become shared as a result
of direct interaction [38–43] and, hence, become parts of group symbolic structure.
Specifically, contagion implies that a sign and/or an association between signs used by
one individual at t1 is reproduced at t2 by another individual socially tied to the first one
(see Fig. 2).
Fig. 2. Socio-symbolic pattern ‘contagion’. Green circles = individuals; green lines = social tie;
blue diamonds = signs; purple line = unshared association between signs; blue line = shared as-
sociation between signs; grey lines = usage of signs.
Fig. 2 represents a situation when a local group member, A, interacts with another group
member, B, and uses the term ‘multi-agency meeting’ that refers to a type of meeting
with officials. At the next point in time, B learns this special term and reproduces the
sign ‘multi-agency’, associating it with the previously known sign ‘meeting’.
3 Technique and Tool to Extract Instances of Symbolic
and Socio-symbolic Micro Patterns
Symbolic and socio-symbolic patterns can be traced in semantic and socio-semantic
networks that contain information on social ties and/or usage of signs and meaningful
associations between them in different social groups
The proposed technique and the pattern retriever tool to infer symbolic and socio-
symbolic patterns involve the construction of semantic and socio-semantic networks
using empirical data; programming of network configurations for symbolic and socio-
symbolic patterns; semi-automatic search, storage, and visualisation of instances of pat-
terns in the networks, and extraction of textual contexts in which the patterns occurred,
for further qualitative analysis.
3.1 Producing Semantic and Socio-semantic Networks
To produce ‘independent’ and ‘dependent’ semantic and socio-semantic networks for
two points in time, we map semantic, social, and sign usage networks from empirical
data that include texts and sociometric surveys.
Semantic networks are produced from textual data based on the co-occurrence of
signs (words) within a certain textual context in sentences. First, the texts representing
a symbolic structure of a group at a certain point in time are combined in a corpus using
the quanteda  package in R. Then, the texts in the corpus are converted into sets of
all grammatical forms of words encountered in the texts, using the UDPipe  pack-
age. The words are tagged with a corresponding part of speech (POS) and then con-
verted into their dictionary forms through lemmatisation. The lemmatised words are
then combined with POS tags (e.g. ‘flood_verb’ and ‘flood_noun’) to allow distinguish-
ing between different meanings of the same word when it is used as a noun, a verb, or
an adjective. We consider these lemmatised words combined with their POS tags as
signs. Additionally, punctuation is removed except for full stops to preserve sentence
boundaries. Then, signs other than nouns, verbs, and adjectives, as well as signs from
a customised stop list, are marked to be omitted from the networks. Finally, the co-
occurrence of signs within a textual context (‘window’) of several signs (unless sepa-
rated by a full stop, i.e. sentence boundary) is counted for each text in the corpus. Note
that the size of the ‘window’ has to be adjusted to optimally capture meaningful
Note that to trace symbolic patterns in the most accurate way, the ‘independent’ symbolic struc-
ture has to be captured at least at a single point in time, and the ‘dependent’ symbolic structure
has to be captured at least at two points in time. This way, it is possible to trace changes in
appearance of signs and sign associations in the ‘dependent’ symbolic structure at t2 compared
to the same symbolic structure at t1 and to the ‘independent’ symbolic structure at t1. While
longitudinal data are desirable, cross-sectional data can also be used.
associations between signs depending on the type and amount of original textual data
[see 44, 45]. These co-occurrence counts are used to produce lists of nodes (signs) and
links (sign associations) that are then converted into semantic networks using igraph
package . The semantic networks are then binarised using researcher-defined
threshold values to retain stable meaning structures. Different binarisation threshold
values can be chosen, depending on a type and size of textual data in a corresponding
corpus. For example, if texts in a corpus contain many complex associations between
signs (e.g. large written texts such as documents), a higher threshold can be chosen. If
texts contain fewer complex associations between signs (e.g. sources from oral speech
such as transcripts of loosely structured interviews), a lower threshold can be used.
After the creation of semantic networks per each text in the corpus, the semantic
networks corresponding to the corpus are combined in a threshold-based merge seman-
tic network. This network includes all links that appear in at least n semantic networks,
where n is a researcher-defined threshold. For our purposes, we create merge networks
using n of 1 and 2. The merge semantic network based on the threshold of 1 includes
signs and associations between signs that occur in one or more semantic networks of
texts in the corpus. This merge network preserves information on exclusive and com-
mon signs and associations between signs in the corpus that is needed to trace the socio-
symbolic patterns. The merge semantic network based on the threshold of 2 contains
signs and associations between signs used in two and more semantic networks of texts
in the corpus. It preserves information only on common signs and associations between
signs and dismisses those used only in a single text as irrelevant or idiosyncratic (this
applies for tracing symbolic patterns). Furthermore, isolated signs in the merge net-
works are deleted.
Social networks are mapped by importing sociometric matrices in R and converting
them into networks using igraph package. Then, the resulting social networks are bi-
Bipartite sign usage networks are produced by creating nodes representing different
texts in the corpora and linking them with the signs used in those texts. Since the texts
can be associated with individuals from the social networks produced earlier, the bipar-
tite network represents individuals and all signs they used.
A socio-semantic network is produced as a union of a social network, a bipartite sign
usage network, and a threshold-based merge semantic network.
Further, to find and extract symbolic and socio-symbolic patterns of the interplay
between different symbolic structures in the context of social ties, we construct com-
bined quasi-longitudinal semantic and socio-semantic networks. To enable the con-
struction of such networks, we have developed a special coding scheme that reflects
usage of signs and associations between signs by two types of actors, as well as the
occurrence of social ties at different points in time.
3.2 Coding Scheme
The coding scheme is used to apply codes to the threshold-based merge semantic and
socio-semantic networks and to interpret the codes in the combined quasi-longitudinal
networks. For illustrative purposes, we describe the scheme that encodes social ties of
local actors, the usage of signs by local actors, and occurrence of associations between
signs in expert and local texts at one and two points in time.
The coding scheme uses three exclusive sets of numerical codes (see Table 1). The
first one is the ‘basic’ set, containing three codes, ‘1’, ‘2’, and ‘5’. Each code in this set
indicates the usage of signs and associations between signs by locals or experts at a
single point in time captured in the locals’ networks at t1 and t2 and the experts’ network
at t1, respectively. The first two codes in the set also indicate occurrence of social ties
at t1 and t2. The second set is ‘cumulative’, containing four codes, ‘3’, ‘6’, ‘7’, and ‘8’.
The codes from the ‘cumulative’ set capture all possible cases of the usage of signs and
their associations in longitudinal data at more than one point in time by more than one
type of actor, as well as the occurrence of social ties. Each code in the ‘cumulative’ set
corresponds to a sum of two or three code values from the ‘basic’ set. For example, the
code ‘3’ is a sum of the values of the codes ‘1’ and ‘2’, which represent locals’ node
and link usage at the first and the second point in time, respectively. Furthermore, each
code in the ‘cumulative’ set corresponds to only one possible combination of code val-
ues from the ‘basic’ set. Thus, one may unambiguously interpret the ‘cumulative’ code
values of nodes and links in the combined quasi-longitudinal networks to find out to
which point(s) in time and to which type(s) of actor(s) the sign usage links and associ-
ations between signs correspond, and at which point in time the social ties occur. In
addition, the third ‘node type’ set contains two codes, ‘10’ and ‘11’, that are used to
designate the types of nodes in the quasi-longitudinal socio-semantic networks.
Table 1. The coding scheme for threshold-based merge semantic and socio-semantic networks.
Codes indicate social ties and local and/or expert usage of signs and associations between them
at the first and/or the second point in time, as well as the type of nodes.
locals at t1
locals at t2
experts at t1
locals at t1 and t2
experts and locals at t1
experts at t1 and locals at t2
experts at t1 and locals at t1 and t2
‘node type’ set
3.3 Constructing a Combined Quasi-longitudinal Semantic Network
As the input for creating a combined quasi-longitudinal semantic network, three thresh-
old-based merge semantic networks are used: ‘independent’ expert network at t1 (based
on the threshold of 1) and ‘dependent’ local networks at t1 and t2 (based on the threshold
of 2). The ‘independent’ semantic network contains common as well as exclusive signs
and associations between signs that are used relatively regularly (i.e. the frequency of
their usage is above a binarisation threshold chosen at the stage of semantic mapping).
The signs and associations between signs in the ‘dependent’ semantic networks are rel-
atively common (used in at least two texts in the corresponding corpus) and relatively
regular (as defined by a chosen binarisation threshold). In all networks, the signs are
associated with at least one other sign (i.e. there are no isolated signs).
Using the igraph package in R, for each merge semantic network, the codes are as-
signed to all signs (as node attributes) and associations between signs (as link values)
according to the developed coding scheme (see Table 1). Then, the networks are con-
verted into data frame format to enable further manipulations. The data frames are com-
bined, automatically summing the assigned code values. Then, the resulting data frame
is converted back into the network format. This procedure results in the combined
quasi-longitudinal semantic network where sign and sign association codes indicate a
point in time at which they were used in the ‘independent’ and the ‘dependent’ symbolic
structures (see the schematic representation of the union procedure in Fig. 3).
Fig. 3. Procedure for creating a combined quasi-longitudinal semantic network. Toy semantic
networks, left to right: ‘independent’ at t1, ‘dependent’ at t1, ‘dependent’ at t2, combined quasi-
longitudinal. Diamonds = signs; red lines = association between signs used only in the ‘inde-
pendent’ network; blue lines = association between signs used only in the ‘dependent’ net-
works; orange line = association between signs used in both types of networks. Letters in node
labels shown for illustrative purposes; sign and sign association codes are shown as node and
3.4 Constructing a Combined Quasi-longitudinal Socio-semantic
As the input for constructing a combined quasi-longitudinal socio-semantic network,
we use ‘dependent’ merge semantic networks based on the threshold of 1 (where signs
and associations between signs are not necessarily common
), sign usage networks and
social networks for t1 and t2.
In R, for each association between signs in each merge semantic network, the igraph
package is used to code whether a particular association between signs occurred in a
particular text from a corpus and hence was used by a particular individual or not. Spe-
cifically, for each association between signs, we add attributes with codenames of all
individuals in a group, such as ‘A’, ‘B’, ‘C’, and fill the attributes with binary values
Considering unshared signs and associations between them allows us to trace their introduc-
tion into group symbolic structure presupposed by several socio-symbolic patterns.
+ + =
5b (1) c (1)
indicating their (non-)usage by corresponding individuals. For example, for an associ-
ation between signs used only by A and C, the attributes ‘A’, ‘B’, and ‘C’ would have
the codes ‘1’, ‘0’, and ‘1’, respectively.
Then, for each point in time, the threshold-based merge semantic, sign usage and
social networks are combined into a socio-semantic network that represents ties be-
tween individuals using particular signs and associations between signs. Next, in each
resulting socio-semantic network, a code is assigned to all social ties, sign usage links,
and associations between signs (as link values), reflecting their usage at a specific point
in time according to the coding scheme (see Table 1).
Finally, the socio-semantic networks at t1 and t2 are combined through a union pro-
cedure, automatically summing the code values. In the resulting network, additional
codes are assigned to all nodes (as node attributes) indicating whether a node represents
an individual (coded as 10) or a sign (coded as 11). This procedure results in a combined
quasi-longitudinal socio-semantic network (see the schematic representation of the un-
ion procedure in Fig. 4).
Fig. 4. Procedure for creating a combined quasi-longitudinal socio-semantic network. Toy so-
cio-semantic networks, left to right: ‘dependent’ at t1, ‘dependent’ at t2, quasi-longitudinal. Dia-
monds = signs; circles = individuals; blue lines = association between signs; green lines = so-
cial tie between individuals; grey lines = usage of signs by individuals. Letters in node labels
shown for illustrative purposes; sign association codes shown as link labels. Node codes reflect
3.5 Programming Configurations for Symbolic and Socio-symbolic
To represent theoretically derived symbolic and socio-symbolic patterns in a computer-
readable format, we programme network configurations for each pattern using igraph
package. Programmed network configurations represent patterns as nodes connected
by links. The nodes and the links in a pattern are assigned codes according to the coding
scheme (see Table 1), based on the theoretical description of a pattern.
In programmed network configurations for symbolic patterns, nodes represent signs
and links represent associations between signs. Consider the example of the configura-
tion for the symbolic loose coupling pattern that concerns the interplay between expert
and local symbolic structures. The pattern indicates that locals reproduce experts’ sign
c associating it with their pre-existing sign a, while ignoring the experts’ association
between c with b (see Fig. 5). The programmed configuration for this pattern consists
of three nodes a, b, and c connected by two links overlapping through the node c. All
information about the usage of signs and associations between signs characteristic of
the loose coupling pattern is reflected in the codes assigned to the nodes and the links
in the programmed configuration according to the coding scheme. For example, the
node c is assigned the code ‘7’ that indicates that experts use the sign c at the first point
in time (coded as ‘5’) and that locals use the same sign at the second point in time
(coded as ‘2’). Another node, a, (coded as ‘3’) corresponds to the sign a that locals use
at both points in time (coded as ‘1’ and ‘2’) while experts do not use at all, and that
becomes associated with experts’ sign c at the second point in time (the sign association
coded as ‘2’). The remaining node and link are coded according to the same logic.
Fig. 5. Programmed network configuration for pattern ‘loose coupling’. Diamonds = signs;
lines = associations between signs. Sign and sign association codes shown as node and link la-
bels. Letters in node labels are not part of the programmed configuration and are shown for il-
In programmed network configurations for socio-symbolic patterns, nodes represent
individuals and signs, and links represent social ties, sign usage by individuals, and
associations between signs. Links are assigned with codes indicating their occurrence
at one or two points in time according to the coding scheme (see Table 1). In addition,
nodes are assigned with codes indicating whether they represent an individual (‘10’) or
a sign (‘11’).
Programmed configurations are stored in R as igraph objects. They will be used to
find instances of patterns in empirical data.
3.6 Finding, Extracting, and Visualising Instances of Patterns
In this section, we describe the pattern retriever, an R script for semi-automatic retrieval
of patterns from network data. The tool allows us to find, extract, and visualise instances
of symbolic and socio-symbolic patterns of the interplay between symbolic structures
in the context of social ties, as well as to find and extract textual contexts of associations
between signs part of the instances of the patterns. The pattern retriever conducts the
1. Network pre-processing. The combined quasi-longitudinal networks are pre-pro-
cessed: links are symmetrised, and multiple edges and self-loops are removed.
2. Pattern retrieval. Using standard igraph functions, the previously programmed net-
work configurations are used as ‘search terms’ to find parts of the combined quasi-
longitudinal semantic and socio-semantic networks that correspond to the symbolic
and socio-symbolic patterns. This operation is implemented as follows. First, for
each pattern, the algorithm looks up parts of the quasi-longitudinal network that
correspond to a programmed configuration for a pattern and extracts a list of nodes
corresponding to the pattern. The difficulty is to extract the links that correspond to
the pattern, given that not all links connecting the extracted nodes in the quasi-lon-
gitudinal network correspond to this pattern. Consider an instance of the pattern a–
c–b found in a quasi-longitudinal network. While the quasi-longitudinal network
contains nodes a, b, and c, it may also contain a link a–b that does not correspond to
the pattern. To retrieve only those links that correspond to a pattern, the algorithm
makes use of link codes stored in the programmed network configurations and the
quasi-longitudinal networks. First, the algorithm extracts all links connecting the
derived nodes, including those that do not correspond to a queried pattern. Then, to
remove unrelated links, the algorithm filters the extracted links based on the code
values of links in the programmed network configuration for the queried pattern.
Finally, the algorithm constructs networks for separate extracted instances of the
pattern and for all its instances, which are stored as separate igraph objects in R.
3. Pattern visualisation. To enable qualitative analysis, visualisations of the extracted
instances of patterns are created. For each pattern, all its instances are visualised in
a single network plot (see example in Fig. 6). These visualisations allow the in-depth
examination of the interplay between symbolic structures as well as between social
and symbolic structures. They enable an understanding of the structural organisation
of patterns. For example, visualisations allow us to identify the most central signs
(that appear in many instances of a pattern) that would have to be subjected to further
Fig. 6. Extracted toy network containing all instances of the ‘loose coupling’ pattern. Dia-
monds = signs; lines = associations between signs. Sign labels and codes are hidden. A single
instance of the pattern is shown in red for illustrative purposes.
In addition, every single instance of a pattern is visualised in a separate network plot
(see Fig. 7). For each pattern, all these visualisations are exported for further qualitative
Fig. 7. A single instance of ‘loose coupling’ pattern. Blue diamond and line = sign and associa-
tion used only in the ‘dependent’ network; orange diamonds = signs used in ‘independent’ and
‘dependent’ networks; red line = association between signs used only in the ‘independent’ net-
work. Sign labels show lemmas, part-of-speech information, and codes; link labels show sign
4. Textual contexts extraction. Finally, to facilitate further qualitative analysis, for
each instance of a pattern, functions in the quanteda package are used to semi-auto-
matically extract all textual contexts containing the associations between signs ap-
pearing in that instance of a pattern from a corresponding textual corpus.
In the next section, we illustrate the application of the technique and the pattern retriever
tool to analyse the interplay between expert and local symbolic structures in the context
of social ties in two local groups engaged in flood risk management.
The source of test data is an ethnographic study of two local flood management groups
in England. The data were collected during six weeks of fieldwork in two villages in
the County of Shropshire, England.
The dataset consists of cross-sectional textual data on expert and local symbolic
structures as well as of sociometric data on relationships between local flood group
members. The data representing expert symbolic structures were collected in the form
of relevant documents (totalling around 316,000 words) produced by official flood risk
management agencies and authorities to inform local groups and other stakeholders
about flood risk management measures, activities, and strategies. Information on local
symbolic structures was collected through semi-structured interviews with 15 members
of the two ‘local flood groups’ voluntarily involved in flood risk management in the
two villages (henceforth, LFG I and LFG II). The corpus of interview transcripts con-
tains around 186,000 words. The data on social relationships, i.e. friendship and col-
laboration, within the local groups were collected using sociometric surveys. The data
were processed as described in the previous section to produce combined cross-sec-
tional semantic and socio-semantic networks
The semantic networks were mapped based on co-occurrence of signs within the window of
9 (i.e. separated by 7 signs) in the texts. Signs from a customized stop list as well as those
with part of speech other than noun, verb, or adjective, were not included in the networks.
4.2 Symbolic and Socio-symbolic Patterns in Two English Flood-prone
Applying the described technique and the tool to our empirical data, we extracted in-
stances of 17 symbolic and 7 socio-symbolic theoretically proposed patterns to manu-
ally confirm their ethnographic relevance. In what follows, we provide illustrations of
such manual evaluation using one symbolic pattern, loose coupling, and one socio-
symbolic pattern, contagion.
As described in Section 2, the pattern loose coupling reflects how a sign from an
‘independent’ symbolic structure is reproduced within a ‘dependent’ symbolic structure
being associated with a pre-existing sign specific to the ‘dependent’ structure. Follow-
ing the procedure described in Section 3, we started with extracting all the instances of
the pattern from the combined semantic network. Then, we visualised all the instances
of the pattern in a network plot. For illustrative purposes, we focus on the instances of
the loose coupling pattern involving one expert sign reproduced by the locals, ‘plan’
(see Fig. 8).
Fig. 8. Instances of the pattern ‘loose coupling’ containing the sign ‘plan’. Red diamonds and
lines = signs and their associations used only by the experts; blue diamond and line = sign and
association between signs used only by the locals; orange diamonds = signs used by the experts
and locals. For illustrative purposes, only labels of signs that appear in the example are shown.
Some signs are hidden to reduce visualisation complexity. Signs’ part-of-speech is hidden.
The visual representation demonstrates that the sign ‘plan’ used by the experts and lo-
cals has many more associations in the expert symbolic structure than in the local one.
This reveals that the meaning of ‘plan’ is more elaborated for the experts than for the
locals. It is, therefore, likely that the associations containing this sign are imposed on
Semantic networks for local symbolic structures contain signs and their associations used at
least two times. Semantic networks for expert symbolic structures contain signs and their as-
sociations used at least eight times.
the locals. For example, the association between ‘plan’ and ‘strategy’ is an expert-spe-
cific association. Meanwhile, the association between ‘plan’ and ‘neighbourhood’ as
well as the latter sign itself are used only by the locals. This means that the locals,
reproducing the sign ‘plan’, discard the association used in the expert’s symbolic struc-
ture with the same sign (‘plan–strategy’) and embed this sign in the local symbolic
structure by creating a new association with their pre-existing sign ‘neighbourhood’.
This can be preliminarily interpreted as the locals’ re-appropriation of the expert sym-
bolic structure through the association between the sign ‘plan’ with the more locally
relevant sign, ‘neighbourhood’.
To put our interpretation of the pattern to test, we extracted and manually examined
all textual contexts that contain the associations between the sign ‘plan’ with the signs
‘management’ and ‘neighbourhood’ from the original official documents (N = 240) and
the interviews with the local flood groups members (N = 9).
The following quote is illustrative of the meaning of the association ‘plan–strategy’
for the experts:
The Department’s capacity-building support for lead local flood authorities has been
well received. It has provided funding support to train staff from across all local
authorities to improve their knowledge and expertise of flooding. In addition, the
Agency has seconded staff to the local authorities to provide additional resource to
complete strategies and develop sustainable urban drainage system plans. This is a
reciprocal arrangement where some local authority staff have also come into the
Agency to improve their understanding of surface water issues. (Expert document,
The experts strive for a systematic and coordinated approach to flood risk management.
‘Plans’ and ‘strategies’ are the instruments they use to ensure flood management activ-
ities of different stakeholders are aligned and exercised in a timely manner. Hence,
words ‘plan’ and ‘strategy’ are firmly ingrained in the experts’ vocabulary.
The analysis of the textual contexts extracted from the interviews with members of
the LFG II reveals a local meaning of the association between ‘plan’ and ‘neighbour-
hood’ that is best illustrated with the following quote:
I suppose… the other one [issue] which isn’t perhaps as major [a problem] but it [is]
certainly significant for [the village], is the local developers. The planning permis-
sions are granted on the understanding that certain flood mitigation steps will be
taken. They’re… not necessarily everything that was promised initially happened.
Developers are only allowed to develop in line with the neighbourhood plan. If
there’s no late neighbourhood plan, then they can come in and develop more or
less all they want. So [the local steering group] set up to develop a neighbourhood
plan and it was very successful in doing that and worked very well. Now our group
fed into the neighbourhood plan. (Informant G, a member of the LFG II)
On the one hand, the locals do reproduce the experts’ sign ‘plan’. This happens when
the LFG II accepts the idea of organising and coordinating flood management stake-
holders’ activities in accordance with a certain scheme of action. Hence, the sign ‘plan’
becomes reproduced in the symbolic structure of the flood group. On the other hand, as
opposed to a general document that regulates collaboration between stakeholders across
a wide range of flood management activities and contexts, the local flood group uses
the sign ‘plan’ when it speaks about coordination of activities between stakeholders
involved in local land planning and development. This is done to ensure that new build-
ings and infrastructure do not adversely impact drainage increasing flood risk in the
village. The necessity for the developers to account for the flood risk in the village is
outlined in a ‘neighbourhood plan’ – a local document guiding planning and develop-
ment in the parish that the LFG II often refers to when it speaks about flood-related
problems in the local area. Hence, the reproduced expert sign ‘plan’ becomes appropri-
ated in the local symbolic structure through the association with the locally relevant
To sum up, our initial interpretation of the loose coupling pattern is supported by the
manual inspection of the textual data, supplemented with our ethnographic knowledge
of the field.
The socio-symbolic pattern contagion involves reproduction of signs and/or associ-
ations between them used by one individual in a group by his or her social network
alter, who has not used them before, so that such signs and/or associations between
them become shared as a result of direct interaction. We extracted all instances of the
pattern from the socio-semantic network of the local flood groups and visualised them.
The visual representation of all the instances of contagion pattern for a specific pair of
interacting individuals is provided in Fig. 9. We focus on one of the instances of the
contagion pattern involving the association between signs ‘multi-agency–meeting’ re-
produced by informants D and E.
Fig. 9. Instances of the pattern ‘contagion’ for individuals D and E from LFG II. Blue dia-
monds = signs; green circles = individuals; blue lines = associations between signs; red line be-
tween individuals = social tie; grey lines = sign usage. Links in a single instance of the pattern
highlighted in red. For illustrative purposes, only labels of signs that appear in the example are
shown. Individuals’ labels contain the codename of the group to which they belong.
The figure shows all signs used by informants D and E who are the members of the
LFG II, as well as corresponding sign associations in the local symbolic structure. Note
that not all sign associations are necessarily used by D or E. We identify sign associa-
tion used by a specific pair of individuals by looking at sign association attributes indi-
cating their (non-)usage by individuals in a group (see 3.4). For instance, this way we
confirm that D uses the signs ‘multi-agency’ and ‘meeting’ and associates them into
the term ‘multi-agency meeting’. Related to D with a direct social tie, the informant E
also uses the same two signs and associates them with each other in the same way as D
does. We assume that in the process of interaction with D, the informant E reproduces
the association between the signs ‘multi-agency’ and ‘meeting’ as used by the inform-
The extraction of the textual contexts for each instance of the association ‘multi-
agency–meeting’ in the original interviews with the informants D and E allows us to
verify if the meaning of this association for each informant is similar, and hence, that
our interpretation that this association is shared by both informants is correct. For in-
stance, for the informant D, ‘multi-agency meeting’ is the format that the flood group
uses to work with the official flood risk management authorities, which is best show-
cased by the following quote:
Arcadis [an engineering consulting company] will now be concepting that drawing
out, but each member of the group at the multi-agency meetings… will say that this
is a problem here and that’s a problem there. Like the area at the back of Beech
Drive, I know from my childhood… it’s been an area of flooding. (Informant D, a
member of the LFG II)
Interacting with D, the informant E – who is a newcomer to the flood group – learns
the very idea of getting the agencies around the table at multi-agency meetings and
reproduces the association between signs ‘multi-agency meeting’:
I’ve had quite a few individual meetings, just me and [another member], we just had
a chat about how things are going and where we need to push things generally. Be-
fore multi-agency meetings, me and [another member] would have a chat. (Inform-
ant E, a member of the LFG II)
Thus, the analysis of the textual contexts confirms our expectation that the association
‘multi-agency–meeting’ that comprises the pattern has a similar meaning for D and E,
which, as we know from the ethnographic work, is likely to result from contagion be-
tween a more senior member and a novice.
This paper has dealt with the lack of techniques and tools to analyse the interplay be-
tween symbolic and social structures at the micro level by inferring symbolic and socio-
symbolic patterns that reflect the usage of signs and their associations by socially tied
individuals in specific practical contexts. Such patterns allow for examining the co-
evolution of symbolic structures of different groups while controlling for the effect of
intra-group social network structures on this process. We introduced a technique and a
software tool for semi-automatic location, extraction, storage, and visualisation of
instances of specific patterns in textual data. We illustrated the technique and the tool
by analysing two patterns of interplay between expert and local symbolic structures in
the context of local social structures using empirical data from our 2019 study of two
local flood management groups in England. We conducted a subsequent qualitative
analysis of the two instances of the patterns to ensure that our interpretation corresponds
to the ethnographic knowledge of the field (see ). The limitation of the present il-
lustration is that these data are cross-sectional. Tests based on more extensive longitu-
dinal datasets are to follow.
Acknowledgements. This work was supported by the Russian Science Foundation
(grant 19-18-00394 ‘Creation of knowledge on ecological hazards in Russian and Eu-
ropean local communities,’ 2019–ongoing). The authors would like to thank two anon-
ymous reviewers for providing valuable comments on an earlier draft of the paper.
1. Basov, N., Breiger, R., Hellsten, I.: Socio-semantic and other dualities. Poetics. 101433
2. Fuhse, J., Stuhler, O., Riebling, J., Martin, J.L.: Relating social and symbolic relations in
quantitative text analysis. A study of parliamentary discourse in the Weimar Republic. Po-
etics. 101363 (2019). https://doi.org/10.1016/j.poetic.2019.04.004
3. Godart, F.C., Galunic, C.: Explaining the Popularity of Cultural Elements: Networks, Cul-
ture, and the Structural Embeddedness of High Fashion Trends. Organization Science. 30,
151–168 (2019). https://doi.org/10.1287/orsc.2018.1234
4. Mohr, J.W., White, H.C.: How to model an institution. Theory and Society. 37, 485–512
5. Padgett, J.F., Prajda, K., Rohr, B., Schoots, J.: Political discussion and debate in narrative
time: the Florentine Consulte e Pratiche, 1376–1378. Poetics. (2020).
6. Schoots, J., Rohr, B., Prajda, K., Padgett, J.F.: Conflict and revolt in the name of unity:
Florentine factions in the Consulte e Pratiche on the cusp of the Ciompi Revolt. Poetics.
101386 (2020). https://doi.org/10.1016/j.poetic.2019.101386
7. Bródka, P., Chmiel, A., Magnani, M., Ragozini, G.: Quantifying layer similarity in multiplex
networks: a systematic study. Royal Society Open Science. 5, 171747 (2018).
8. Roth, C., Cointet, J.-P.: Social and semantic coevolution in knowledge networks. Social
Networks. 32, 16–29 (2010). https://doi.org/10.1016/j.socnet.2009.04.005
9. Basov, N., Lee, J.-S., Antoniuk, A.: Social Networks and Construction of Culture: A Socio-
Semantic Analysis of Art Groups. In: Cherifi, H., Gaito, S., Quattrociocchi, W., and Sala,
A. (eds.) Complex Networks & Their Applications V. pp. 785–796. Springer International
Publishing, Cham (2017)
10. Linzhuo, L., Lingfei, W., James, E.: Social centralization and semantic collapse: Hyperbolic
embeddings of networks and text. Poetics. 101428 (2020). https://doi.org/10.1016/j.po-
11. Blumer, H.: Symbolic Interactionism: Perspective and Method. University of California
12. Mead, G.H.: Mind, Self and Society from the Standpoint of a Social Behaviorist. Chicago
University Press (1934)
13. Basov, N.: The ambivalence of cultural homophily: Field positions, semantic similarities,
and social network ties in creative collectives. Poetics. (2019). https://doi.org/10.1016/j.po-
14. Basov, N., de Nooy, W., Nenko, A.: Local meaning structures: mixed-method sociosemantic
network analysis. Am J Cult Sociol. (2019). https://doi.org/10.1057/s41290-019-00084-9
15. R Core Team: R: A language and environment for statistical computing. (2013)
16. Carley, K.: Extracting culture through textual analysis. Poetics. 22, 291–312 (1994).
17. Hardy, C., Maguire, S.: Organizing Risk: Discourse, Power, and “Riskification”. Academy
of Management Review. 41, 80–108 (2016). https://doi.org/10.5465/amr.2013.0106
18. Mohr, J.W.: Soldiers, mothers, tramps and others: Discourse roles in the 1907 New York
City charity directory. Poetics. 22, 327–357 (1994). https://doi.org/10.1016/0304-
19. Fine, G.A.: Group Culture and the Interaction Order: Local Sociology on the Meso-Level.
Annual Review of Sociology. 38, 159–179 (2012). https://doi.org/10.1146/annurev-soc-
20. Puzyreva, K., Basov, N.: Local Knowledge in Russian Flood-Prone Communities: A Case
Study on Living with the Treacherous Waters. In: Babu, G. and Qamaruddin, M. (eds.) In-
ternational Case Studies in the Management of Disasters. Emerald Publishing Limited
21. Hoetker, G., Agarwal, R.: Death Hurts, But It Isn’t Fatal: The Postexit Diffusion of
Knowledge Created by Innovative Companies. Academy of Management Journal. 50, 446–
467 (2007). https://doi.org/10.5465/amj.2007.24634858
22. Binder, A.: For love and money: Organizations’ creative responses to multiple environmen-
tal logics. Theor Soc. 36, 547–571 (2007). https://doi.org/10.1007/s11186-007-9045-x
23. Fiol, C.M., O’Connor, E.J.: Waking Up! Mindfulness in the Face of Bandwagons. AMR.
28, 54–70 (2003). https://doi.org/10.5465/amr.2003.8925227
24. Schilke, O.: A Micro-institutional Inquiry into Resistance to Environmental Pressures.
Academy of Management Journal. 61, 1431–1466 (2018).
25. Nenko, A., Khokhlova, A., Basov, N.: Communication and knowledge creation in urban
spaces: The tactics of artistic collectives in Barcelona, Berlin and St. Petersburg. In: Aiello,
G., Tarantino, M., and Oakley, K. (eds.) Communicating the City. Peter Lang, Austria
26. Hallett, T.: The Myth Incarnate: Recoupling Processes, Turmoil, and Inhabited Institutions
in an Urban Elementary School. American Sociological Review. 75, 52–74 (2010).
27. Joseph, J., Ocasio, W., McDonnell, M.-H.: The Structural Elaboration of Board Independ-
ence: Executive Power, Institutional Logics, and the Adoption of CEO-Only Board Struc-
tures in U.S. Corporate Governance. Academy of Management Journal. 57, 1834–1858
28. Bolton, C.D.: Some Consequences of the Meadian Self. Symbolic Interaction. 4, 245–259
29. Etzrodt, C.: The Foundation of an Interpretative Sociology: A Critical Review of the At-
tempts of George H. Mead and Alfred Schutz. Hum Stud. 31, 157–177 (2008).
30. Basov, N., Brennecke, J.: Duality Beyond Dyads: Multiplex Patterning of Social Ties and
Cultural Meanings. In: Groenewegen, P., Ferguson, J.E., Moser, C., Mohr, J.W., and Bor-
gatti, S.P. (eds.) Research in the Sociology of Organizations. pp. 87–112. Emerald Publish-
ing Limited (2017)
31. Fuhse, J.: The Meaning Structure of Social Networks. Sociological Theory. 27, 51–73
32. Godart, F.C., White, H.C.: Switchings under uncertainty: The coming and becoming of
meanings. Poetics. 38, 567–586 (2010). https://doi.org/10.1016/j.poetic.2010.09.003
33. Rawlings, C.M., Childress, C.: Emergent Meanings: Reconciling Dispositional and Situa-
tional Accounts of Meaning-Making from Cultural Objects. American Journal of Sociology.
124, 1763–1809 (2019). https://doi.org/10.1086/703203
34. White, H.C.: Identity and control: A structural theory of social action. Princeton University
Press, Princeton, N.J. (1992)
35. Antonyuk, A., Puzyreva, K., Basov, N.: Principles and patterns of interaction between expert
knowledge and local community knowledge. Manuscript in preparation, Centre for German
and European Studies, St. Petersburg, (2020)
36. Antonyuk, A., Kretser, I., Basov, N.: Creation of local knowledge in interaction. Un-
published manuscript, Centre for German and European Studies, St. Petersburg, (2019)
37. Vadera, A.K., Aguilera, R.V.: The Evolution of Vocabularies and Its Relation to Investiga-
tion of White-Collar Crimes: An Institutional Work Perspective. J Bus Ethics. 128, 21–38
38. Burt, R.S.: Social Contagion and Innovation: Cohesion versus Structural Equivalence.
American Journal of Sociology. 92, 1287–1335 (1987)
39. Carley, K.: Knowledge acquisition as a social phenomenon. Instr Sci. 14, 381–438 (1986).
40. Coleman, J.S.: Social Capital in the Creation of Human Capital. American Journal of Soci-
ology. 94, S95–S120 (1988). https://doi.org/10.1086/228943
41. Monge, P.R., Contractor, N.S.: Theories of Communication Networks. Oxford University
42. Zhou, D., Ji, X., Zha, H., Giles, C.L.: Topic Evolution and Social Interactions: How Authors
Effect Research. In: Proceedings of the 15th ACM International Conference on Information
and Knowledge Management. pp. 248–257. Association for Computing Machinery, New
York, NY, USA (2006)
43. Cucchiarelli, A., D’Antonio, F., Velardi, P.: Semantically interconnected social networks.
Soc. Netw. Anal. Min. 2, 69–95 (2012). https://doi.org/10.1007/s13278-011-0030-z
44. Benoit, K., Watanabe, K., Wang, H., Nulty, P., Obeng, A., Müller, S., Matsuo, A.: quanteda:
An R package for the quantitative analysis of textual data. The Journal of Open Source Soft-
ware. 3, 774 (2018). https://doi.org/10.21105/joss.00774
45. Straka, M., Straková, J.: Tokenizing, POS Tagging, Lemmatizing and Parsing UD 2.0 with
UDPipe. In: Proceedings of the CoNLL 2017 Shared Task: Multilingual Parsing from Raw
Text to Universal Dependencies. pp. 88–99. Association for Computational Linguistics,
46. Csardi, G., Nepusz, T.: The igraph software package for complex network research. Inter-
Journal Complex Systems. (2006)