nueva página del texto (beta)
Inglés (pdf)
Artículo en XML
Referencias del artículo
Traducción automática
Enviar artículo por email
Citado por SciELO 
Similares en
SciELO 
Permalink
Introduction. A narrative source for the Norman kingdom of Sicily
One of the most important sources for the study of the kingdom of Sicily, the Liber de Regno Sicilie, attributed to ‘Hugo Falcandus,’ provides a vivid and detailed account of the Norman kingdom of Sicily under William I and the first years of William II (1154-1169). This narrative source has become, for better or worse, a pillar for the understanding of the kingdom of Sicily in the second half of the twelfth century.
The studies introducing both the most recent translations of the Liber present a comprehensive background to an initial examination of the text: whereas Loud and Wiedemann cover the historical ambiance of the text, and provide an overview of the debate (up to 1999) about the authorship of the Liber, as well as a philological exploration on the classical resonances in the text,1 E. Türk offers a short summary of the possible identification of the author, and provides synthesised individual backgrounds on relevant historical characters attested in the Liber: Roger II, William I, Margaret of Navarre, Maio, Matthew of Salerno,2 the bishop-elect Richard, Stephen of Perche, Archbishop Walter of Palermo, and Romuald of Salerno.3 Additionally, M. Fuiano’s treatise on ‘Hugo Falcandus’ has been resourceful for a broad understanding of the morphology, style and intertextuality of the Liber; although slightly out-dated, Fuiano’s text offers a comprehensive study of ‘Hugo Falcandus.’4
The name ‘Hugo Falcandus’ appears for the first time in a printed edition published in Paris in 1550.5 None of the surviving medieval manuscripts cite any authority by the name of ‘Hugo Falcandus’, and no such person can be identified in the surviving twelfth-century charters. However, there seems to be a general agreement among previous and current scholarship that although Falcandus was not necessarily one of the main characters in the narrative, he was certainly a contemporary and probably an eyewitness.6 Nevertheless, the question of the authorship is mostly a technical one, of little relevance for the present study, and probably incapable of solution.
This research project sought an integrative approach that would allow the comprehension of social process present in a textual, narrative source. The study was founded on the interactions between social actors that were involved in the reported machinations, as narrated in the text. I assumed that a relational approach could contribute to the understanding of narrative sources, and the main aim is twofold. First, to design a method for extracting relational data, in order to construct networks that represent the information contained in a narrative source such as the Liber. Second, to explore the practical implications of interpreting the networks thus constructed.
The challenge of how to translate words into numbers, and then back into words became as a result the main focus of my research. In order to focus on the information on social and political processes embedded in the text, I needed to transform the rhetoricised reports into a relational dataset; it was necessary to place the relations, not the individuals, at the centre of the study. The first requirement of such an attempt was to present the process of translating a textual structure into a sociological construct, namely, a socio-relational dataset. After the whole Liber was rewritten through Franzosi’s Quantitative Narrative Analysis, I obtained a dataset that provided a series of narrative sociomatrices ready to be parsed through network analytical tools. Of these, centrality measures turned out to be one of the most fruitful approaches for understanding the entire text’s social dimensions.
Methodology. Narrative network analysis
A mew method for the interpretation of narratives
The central idea of this methodological attempt is that narrative analysis yields an understanding of social relations as embedded in the text, and that the meaning of individuals and communities in a text is conditional on their position in a system of social interactions constructed by the author. One should not only recover facts from the narrative, but also find a way to make sense of it. It is here where a structural and relational analysis seems to offer a useful approach. The narrative analysis on ps. Falcandus’ Liber helps to reveal not only its linguistic properties-a task perhaps better left in the hands of the linguists-, but also the considerable amount of sociological information present in the narrative.7 In Laumann’s words, the peculiar and distinctive feature of network analysis “is to explain, at least in part, the behaviour of network elements-i.e. the social actors-and of the system as a whole by appeal to specific features of the interconnections among the elements.”8 In approaching the author’s perspective, I needed a method that would allow me not to solely gauge its ideological stand or its linguistic resonances, but to explain the social information it contains deriving from the structure of the text. Arising from the toolbox of structural analyses of narratives, the research tool that proved helpful for this endeavour was Quantitative Narrative Analysis (QNA).
QNA is a method designed by Roberto Franzosi to study the behaviour of historical actors as reported by narrative sources.9 This approach to historical narratives, which takes advantage of the invariant structural linguistic properties of any narrative, allows historians and social scientists to study the actions and relations of actants.10 QNA is founded on Halliday’s action-verbal clause,11 which is defined as a process that essentially consists of three components: 1) the process itself; 2) the participants in the process; 3) the circumstances associated with the process.12 In this way, the action-verbal clause as the characteristic narrative process implies interactions among participants. The action-verbal clauses are thus the skeleton of the language of the narrative, and the raw material for identifying the behaviours and interactions of the actants in a text.
After rewriting the entire Liber de Regno Sicilie through QNA, I obtained a dataset that divides the social information embedded in the narrative into 420 events. The total number of attested social interactions, coded as semantic triplets, is 1174, together with 89 social relationships also explicitly attested in the narrative. Each interaction and relationship defines an edge that connects two characters. This dataset thus provides a list of social actants and a list of edges. The different social interactions present in the narrative were labelled in 16 different categories. These were classified, filtered, and then grouped into two general types of interactions: communication and influence.
The narrative’s dataset is therefore used to build up network data stored in two separated adjacency matrices: one for each type of interaction. This means that the list of nodes, interactions, and relationships needs to be transformed into sociomatrices-i.e. an adjacency matrix that offers in a tabular format the data representing interpersonal connections. The network data is not dichotomised, as it is given in valued links (namely, the links can bear a value bigger than 1), depending on how many times each pair of characters interacted in each layer.
Centrality and Prestige
Centrality measures were primarily designed, from graph theory to social network analysis, for the identification of the “more prominent” actant in a network by measuring properties of node location in the network. The degree of importance or prominence of an actant’s position can be derived from the structural properties of his/her specific location within a network, for some positions turn out to be more ‘strategic’ than others. A location in the system can be considered ‘prominent’ if both its adjacent links and the indirect paths that lead to and go through it make that position particularly visible in comparison to other positions in the network. As proposed and formulated by D. Knoke and R. Burt, the extent of the prominence of a position in a system can be measured with two classes: centrality and prestige.13
As also pointed out by D. Knoke and R. Burt, the measure of centrality is well suited to dealing with sociological concepts such as access to and control of information.14 Especially in two of my social-interaction types, communication and influence, social actants are involved in processes of access or control of either people or information; for this reason, centrality may be an appropriate and relevant measure for communication and influence layers. In this way, the most ‘central’ actants in the communication and influence layers are likely to stand for those who have the most access or are the most active.
Both layers contain directed ties, and so the measure of centrality will indicate not just the involvement of an actant among a set of ties, but also a source for such ties.15 Hence, two of the centrality indices may be easily applied and are relevant to the analysis of directional relations: degree and closeness.16 These indices measure the actants as sources of interactions. The degree centrality index of an actant, for directed networks, focuses on the ‘out-degree,’ namely the number of ties that originated from it; whereas an actant’s closeness index indicates the average of the shortest paths from the given actant to all other actants in the network. In other words, by degree centrality, the more central an actant is, the more connections it had. Conversely, by closeness centrality, the more central an actant is, the closer it is to all other actants. As such, closeness is measured using the following formula:
where d (y, x) is the distance between actant (node) y as a target, and actant (node) x as a source. The index derived from this measure is a simple normalisation of ratings, adjusting the obtained values to a notionally common scale between 1 and 0.
On the other hand, prestige focuses on the direction of links leading to an actant; defining “a prestigious actant as one who is the object of extensive ties.”17 In simpler words, prestige is a measure of passive centrality in directed networks, as it focuses on the actant as a recipient of relations, not as a giver. Although the concept of ‘prestige’ may have a positive connotation, this is not always the case, for, in the present context, the term merely denotes the active position of an actant as recipient. For example, in the network layer of enmity interactions, a ‘prestigious’ actant would not be held in high regard by the rest of the characters in the narrative, but would instead emerge as an object of offences and other negative interactions. By measuring prestige, I can identify the most important recipients at the relevant layers of interactions: influence and communication.
For analysis, prestige can be measured in indices of degree and proximity. As is the case with measuring centrality, the degree index of an actant indicates the number of ties. However, it focuses on the actant not as a source, but as a recipient. Thus the degree index for measuring prestige takes the ‘in-degree’ instead. In a similar way, the proximity index is like a ‘closeness measure’ that focuses on distances to rather than from each actant.18 That is, I employ the same formula for closeness centrality, but d (y, x) is understood instead as the distance between actant (node) y as a source, and actant (node) x as a target.
Alternative, more elaborate measures such as Eigenvector Centrality and Page Rank may seem very promising, but epistemologically they present severe issues, at least when dealing with historical and social network analysis. Although they are measures which explore the topological implications of the network in depth, they require making assumptions about the social system they represent. We must be careful with the way in which the algorithm for statistical or structural evaluation is assumed in advance to be how the social process unfolds. On top of this, the database for my historical research is not sufficiently robust for a topological statistical measure to be relevant, especially compared to absolute centrality measures. Furthermore, and more specifically, Eigenvector centrality should only be used in undirected graphs. The Eigenvalues of an asymmetric matrix will be expressed not only as real numbers but also as imaginary ones, and consequently the Eigenvector of a directed graph will not always yield valid results.19
Consequently, and in order to measure centrality and prestige, I calculate first the actant’s out-degree and closeness centrality. Since both the communication and influence layers display characters that did not engage in this specific type of interaction, neither as sources nor as recipients (i.e. isolates in a system), the system representing each interaction type is not strongly or unilaterally connected. Therefore, it is impossible to calculate the closeness index for the total set of characters. I thus excluded those actants with out-degrees equalling 0, and then calculated the normalised values for the reduced but unilaterally connected network.
The first network layer analysed is influence. The entries in this layer’s sociomatrix are the interactions attested in the narrative pertaining to activities of advice, request, order, persuasion and instigation. Characters are thus central if they exercise any influence on others, as in the same way characters are ‘prestigious’ if they are constantly influenced in any way. The second layer to which centrality measures seem to be pertinent is communication. Under this type, social actants are involved with each other in processes of access to or control of information and its spread. The entries in this layer’s sociomatrix are the interactions pertaining to activities of consultation, notification, divulgation, and reply. Social actants are ‘central’ in this layer if they constantly communicate with the rest of the social actants, as in the same way character actants are ‘prestigious’ here if they are the objects of several processes of communication. As in the influence layer, I have excluded the actants with an out-degree of 0 from the sociomatrix and produced a unilaterally connected graph. Lastly, I turn to the characters’ prestige measures.
Further and distinct analysis has been conducted on this multi-layered data, such as the application and assessment of community-detection algorithms, structural equivalences and positional analysis. These other methods and results can be read in full in my 2013 MA thesis.20
Results and interpretations. The analysis of the narrative’s centrality measures
The centrality indices of the influence network layer are shown in Table 1.21 Next is the calculation of the prestige indices; which are shown in Table 2.22 In both tables, the first two columns display the network measures on which the indices’ values are based, and the last two columns display the prestige indices of each property by means of simple normalisation of ratings. In the same way, tables 3 and 4 display the calculated measures and indices of the unilaterally directed network layer of communication.
Table 1 Centrality indices for the influence layer of the network (calculated by dropping the actants with out-degree=0 from the original actants set)
Table 2 Prestige indices for the influence layer of the network (calculated by dropping the actants with in-degree=0 from the original actants set)
At first glance, one can notice several ‘central’ characters. In order of decreasing degree centrality, Stephen of Perche23 is in first place, Maio of Bari24 in second, followed by Queen Margaret,25 then King William I in fourth, and so on. The least central actants in this index, and hence the less demanding or influential, are a numerous group that includes 27 characters, such as William of Lesina,26 pope Alexander III,27 and young King William II. The closeness centrality index shows a different ordering, with the notary Matthew28 in first place, Stephen of Perche and Margaret in joint second, followed by William I and then bishop-elect Richard29 in fourth. The ordering of the less central characters resembles the degree centrality more strongly, a considerable exception being Matthew Bonellus,30 who bears a low closeness index although having a not so low degree centrality.
This difference between the two centrality indices can be explained as the difference between simply influencing a lot of other characters (as indicated by the out-degree), and being able to influence people who are in turn influencing a considerable amount of the rest of the characters (as indicated by the shortest paths that have that specific actant as a source). This implies, for example, that the notary Matthew, although not engaging in many interactions, is influencing actants who affect in turn many other members of the same network. Thus, he is able to indirectly influence those to whom he is not directly tied. Although Stephen of Perche presents a very high number of interactional engagements, and Maio is very heavily linked to another central character, William II, the notary Matthew is still the more central character by closeness, and so can reach more actants in the social system in fewer steps. What is important here is not how many other characters a character is adjacent to, but the ‘centrality’ of those characters. On the other hand, Maio falls from the top central actants when observing his closeness centrality index, suggesting that such a recurrent character in the narrative, highlighted as a major figure of political instigation, may have been less effective as a prominent actant of influence as embedded in the text’s own social milieu.
Measuring centrality allows for the identification of Stephen of Perche, Margaret, and William II as the most influential actants, both by their direct interactions and by the people they influence. Maio and the notary Matthew are also confirmed as crucial characters for understanding the process of exerting influence in the textual source, but with considerable structural limitations, making them less reliable central characters than could be expected. All these distinctions can be visualised embedded in the whole network layer in Figure 1.
Figure 1 Graph representing the influence layer of the network and its centrality (calculated by dropping the actants with out-degree=0 from the original actants set). The node’s size is proportional to its out-degree, and the darkness of its colour is proportional to its closeness centrality index. The link’s width is proportional to its weight (frequency of the interaction), and the arrowhead indicates the target of the directed interaction.
From examining Table 2, one can notice that the degree prestige index covers a wide range of values, displaying a power-law distribution. A power law is a functional relationship between two quantities, where one quantity varies as a power of another. Accordingly, the demonstration of a power-law relation suggests the existence of specific kinds of mechanisms that could underlie the observed phenomenon, and can thus be used to hypothesise on a meaningful connection between proportion and distribution within a social system. In other words, in relation to the present study, a rather minimal number of characters in the Liber de Regno Sicilie were influenced by a vast majority of social actants, whereas this majority of influenced characters were only targeted by a scarce number of interactions. As can be expected, the most active target of influential interactions was the king, William I (with an in-degree=24 and holding the maximum degree prestige index). It is clear that if a social actant wanted to achieve something, or wanted to get something done, the king was unsurprisingly the actant to influence. In second position is Stephen of Perche (with a degree index of .833), making him a very ‘prestigious’ character when it comes to influencing someone. This position clearly highlights Stephen, and suggests a differentiation from other influential actants in the story. Stephen, after being appointed by the Queen Regent as chancellor of the kingdom, emerges not only as a central character in the exercise of influence, but also as a character that other social actants recurrently approach. Considering the development of the narrative events, it seems that the social characterisation at the influence layer of William I was taken over by Stephen of Perche. As is confirmed by the prestige measures, it was not the underage king (William II) or the Queen, as one might have expected in view of her position as Regent, but the newly appointed chancellor who, after the death of King William I, became the active target for the rest of the influential social actants of the royal court. Stephen is followed by Margaret and Matthew Bonellus, who, in spite of having a high degree index (.583 and .542 respectively), still rank significantly lower than him.
The proximity prestige index indicates on the influence layer to what extent the actants who are influencing the specific actant are themselves being ‘influenced.’ Thus this prestige index may be useful for revealing the indirect manipulation a character went through, as is displayed in Table 1 under the label “Normalised Closeness Centrality.” The descending order of the values in this column differs significantly from the values of the degree index. As in the case of the latter, the characters are still distant from each other in terms of prestige, although the range of the proximity index is smaller. Archbishop Hugh, the Lombards,31 and Maio of Bari occupy the top three positions (with a proximity prestige index of 1). This means that these three characters are directly tied to all the people involved in the chain of influence that targeted them. Whoever influenced these three characters was not influencing them as an ‘intermediary.’ Hence these actants are not only unprestigious actants for influence processes, having been scarcely addressed in this layer, but are also characters less prone to manipulation, without having to be omitted from the influence layer as targets too. The value of the proximity index drops considerably (from 1 to .667) in the following place occupied by Simon,32 which is then followed by Margaret (with .457). As the values of the proximity index drop, the characters are presented as under the influence of actants who are in turn being influenced themselves. Thus, the influenced characters with the lowest prestige, but still greater than 0, are placed in a more vulnerable position than the most influential characters that may hide behind a chain of influence. Characters such as Pope Alexander III and Bishop Gentile of Agrigento33 are hence placed in the narrative as social actants at a very disadvantageous place and are susceptible objects of manipulation, since people far from their reach indirectly influence them. The graph that represents the figures of Table 2 is shown in Figure 2.
Figure 2 Graph representing the influence layer of the network and its prestige (calculated by dropping the actants with out-degree=0 from the original actants set). The node’s size is proportional to its in-degree, and the darkness of its colour is proportional to its proximity prestige index. The link’s width is proportional to its weight (frequency of the interaction), and the arrowhead indicates the target of the directed interaction.
Based on the degree centrality index, the most central character is Stephen of Perche, followed by Matthew. The gap between these two positions is slightly larger than the gap between the top two positions of the influence layer. This gap seems to indicate that the chancellor, Stephen of Perche, has a more active role in the informative processes and that he is a central actant in terms of influencing processes. Considering both theinfluenceandcommunicationlayers, the narrative’s social construction presents Stephen of Perche as an actant relying on a different strategy for collaborating with the other members of the royal court, based on influence and communication. This is opposed to the strategies of Maio of Bari, or King William I, who did not appear to be actively involved in communicative interactions.
The position of Matthew Bonellus is also noteworthy; although presented as a less influential character, his characterisation as a major social actant is given more importance in terms of his ability to communicate. I shall return to him when dealing with the closeness centrality index. Gilbert of Gravina occupies the third position by degree centrality,34 another less influential character who becomes much more central in terms of his active characterisation as an actant engaged in many communication processes across the Sicilian narrative. Fourth place is taken by three characters, William I, Maio, and Hugo,35 followed by Margaret, the notary Matthew, Everard,36 and Romuald of Salerno.37
As displayed in Table 3, the closeness centrality index values present again a considerably different order than was the case with the degree centrality, which is even more surprising than the one revealed at the influence layer. Robert of S. Giovanni38 holds the largest value in the closeness index, but in return holds a low degree centrality (.2). Even though Robert was characterised as a less recurrent actant, namely a character that communicated with other characters only on a handful of occasions, he nevertheless can reach almost all remaining communicating actants in no more than two steps. Robert of S. Giovanni thus might not have been informing a lot of people, but he was in touch with people that could transmit the information to a large proportion of the remaining social actants. Robert is followed by Stephen of Perche, as can be expected from a character with such a large degree centrality. In turn he is followed by two unexpected central characters: the notary Matthew, and Odo39 (with a closeness index of .432 and .408, respectively). All of these distinctions embedded in their network layer are visualised in Figure 3.
Table 3 Centrality indices for the communication layer of the network (calculated by dropping the actants with out-degree=0 from the original actants set)
Figure 3 Graph representing the information layer of the network and its centrality (calculated by dropping the actants with out-degree=0 from the original actants set). The node’s size is proportional to its out- degree, and the darkness of its colour is proportional to its closeness centrality index. The link’s width is proportional to its weight (frequency of the interaction), and the arrowhead indicates the target of the directed interaction.
As in the influence layer, the notary Matthew is here vindicated again as a character that indirectly reaches a larger number of actants in very few steps. The case of Robert of S. Giovanni and Odo is even more striking because both their degree centrality and their direct appearance in the narrative suggest they are secondary characters, not at all central to the events reported. However, the location of the social actants with whom Robert and Odo communicate makes both of them prominent actants in the social construct of the author. I hence infer that the Liber’s author tacitly presented Robert of S. Giovanni and Marshal Odo as highly relevant communicative actants. Even though they are not presented as leading characters in the plot, these characters were structurally placed in a strategic position as social actants in his story. It may not be incidental that precisely Robert, a notary, was cognitively described in very positive terms as “a man of high reputation and proven faithfulness, whom no party of conspirators […] No flattering good fortune had ever raised him to a point where his innate good nature was adulterated by any pride or insolence; and no bad fortune had ever cast him down to the point where he wished to hawk his loyalty for sale in exchange for the favour of powerful men or for the value of any honour.”40 It is also important to remember that Robert of S. Giovanni has been suggested before as one of the possible identities of ‘Hugo Falcandus’ by C.A. Garufi,41 relying on previous theories by C.H. Haskins,42 and E. Besta43 that identified the author of the Liber as a royal notary. However, as mentioned above, this theory has been contested.44
Following a pattern similar to the influence layer (as displayed in Table 2), the distribution of the in-degrees of the communication layer (as shown in Table 4) indicates that there are a handful of characters who concentrate the information received from other characters, whereas a significant majority of the communicative characters are only the recipients of information from a single actant, or none at all. Stephen of Perche is placed as a more prestigious actant than former King William I, and much more prestigious than his contemporary ruler, Margaret. Stephen is undoubtedly the largest recipient of information across the social construct of the narrative, since the second most prestigious characters at the communication layer follow him with a degree index of .588. William I and Matthew Bonellus jointly occupy these places, closely followed by Maio (with an index degree of .529). This ordering also resembles the one presented by the degree centrality index, with some minor permutations and the exception of Archbishop Hugh, a character that, despite being an important communicative actant, is presented as an unpopular target for actions of communication. However, the larger in-degree values are even larger than the top out-degrees, pointing out that the most important actants tend to be informed more and be informing less, a strategy that becomes clearer by comparing their centrality and prestige as social actants engaged in communication interactions.
Table 4 Prestige indices for the communication layer of the network (calculated by dropping the actants with out-degree=0 from the original actants set)
The proximity prestige index reveals less insightful information here than the one provided at the influence layer. The communication proximity index indicates how many intermediaries the communication processes, namely the interactions that implied any sort of communication, went through before reaching a certain character. This means that the characters with a proximity index of 1 were engaged by characters that had not been targeted by people outside of their immediate circle of communicative acquaintances. Thereby the characters with a low proximity index are the recipients of information provided by people that had been engaged in communication with other people in turn. So, what the proximity index indicates here is the extent to which an actant is susceptible to receiving information which derives from outside his/her direct contact social reach; thus expanding the character’s range of communication sources. Hence, it is relevant to highlight that William I and Matthew Bonellus, both of whom have a high degree prestige index and a low proximity index, are characters set in a favourable position for receiving information from a wider social scope. These indices are visualised in Figure 4.
Figure 4 Graph representing the information layer of the network and its centrality (calculated by dropping the actants with out-degree=0 from the original actants set). The node’s size is proportional to its in- degree, and the darkness of its colour is proportional to its proximity prestige index. The link’s width is proportional to its weight (frequency of the interaction), and the arrowhead indicates the target of the directed interaction.
Conclusion. Historical implications of a network interpretation
After conducting this experiment on narrative, several points must be made. Depending on what one seeks to focus on, or to find out from a text, one’s approach should provide clearly defined tools for conducting research in systematic ways. Hence, the first challenge I faced was not only finding the names of people and their attributes as depicted in the text and also as attested in other contemporary sources (a task already done laudably by many historians), but also finding information, both explicit and implicit, on how the message presents and depicts people and their social space. What I initially imagined as my object of study, the dynamics of the Sicilian royal court, necessarily became the analysis of narrated, and thus constructed, interactions of its actants. In order to focus on the information about social and political processes embedded in the text, I needed to transform the rhetoricised information into a relational dataset; it was necessary to place the relations, not the individuals, at the centre of the study. The first requirement of such an attempt was to present the process of translating a textual structure into a sociological construct, namely, a socio-relational dataset. The structural foundation and quantitative approach of R. Franzosi’s narrative analysis provided the means of doing so; allowing me to present my reading of the textual source in a transparent and consistent scheme of textual interpretation. As my criteria for reading the text are explicitly stated and theoretically justified, the reader can follow the process of interpretation almost step-by-step. This is useful because of the possibility of tracing errors in the theoretical framework, and of criticising the specific principles employed during my reading; my method may thus provide a platform that can be replicated for readings of other narrative sources. I hope that my database established a controlled field for comparison of information embedded in narratives. This could be the first step into a larger enterprise of creating multiple social constructions, either for constructing a multi-voiced perspective on a specific social reality, or for integrating this social reality into a larger historical picture.
In addition to offering a transparent method that places the relations in the narrative at the centre of my reading, the present attempt also explored the possible analytical tools that can be used to interpret the social relational data contained in the narrative structure. Social network analysis provides a large and diverse number of tools for parsing relational data-as long as the selected tools are suited to the characteristics and limitations of the network data. Experimenting with a few of these tools yielded insights into a narrative social space. As in any other case, the key to conducting a relevant network analysis is having a clear research question and a theory that explains the suitability of the analytical tool for the specific substance of the data to be analysed.
Although not all the resulting measures of network analysis proved to be useful nor contributed interesting insights for understanding the Liber’s implications, the overall results are a valuable addition to the perspective of the source, and are particularly useful for providing more nuanced images of the text. As illustrated above, measures of centrality and prestige proved useful when exploring the narrative interactions of influence and communication. For example, I identified the prominence of Robert of S. Giovanni even though he was characterised as a less recurrent actant (namely a character that communicated with other characters only on a handful of occasions). Yet from the source’s perspective, Robert of S. Giovanni is a highly relevant communicative character, connected with characters that could transfer information to a large proportion of the remaining social actants of the author’s royal court. I also recognised the implicit depiction of Stephen of Perche’s particular strategy for collaborating with other members of the royal court, based not only on influence, but also on communication, as opposed to the strategies of Maio of Bari and King William I, who did not appear to be actively involved in communicative interactions according to the author.
King William I, Matthew Bonellus, or any other protagonist in any historiographical account, are not the centre around which the narrative makes sense, but simply nodal points embedded in a social system that involves many other characters. Their actions play on different layers through which the narrative unfolds, offering deeper information than the morphology, style, and attributional information may suggest. In the short term, at the event level, actants appear as the makers of relations, but in the long run, at the level of the entire narrative structure, the relations are what indeed determine and make the actants in the text, for it is the entire system of relations that determines the topology of the narrative’s social space. The narrative is then revealed as a repository of social relations built up by narration across time.
Networks are not only phenomenological realities, but, as pointed out buy H. White, also “measurement constructs.”45 In this way, narrative analysis, which is also concerned with the social space construed by the narration, can benefit from adopting network-based approaches. Through a relational approach, one can bridge the gap between cognitive and structural standpoints, and hence advance towards an understanding of the social images that lie between the lines.
Therefore, I am convinced that if one intends to use a narrative message as a source, one should consider, as one already considers other features such as intertextuality and style, the implications and significance of the social space that the author himself is constructing though narration. To restrict our vision of narrative sources to just their explicit content and formal features is to be partially blind to the complexity of the narrative that one can grasp from taking a number of points of view.
