Network analysis in legal history: an example from the Court of Friesland

This article focuses on the references (allegations) made by the lawyers in a selected number of cases to Roman and customary law as well as to court decisions when arguing their case. The analysis focuses on three similar civil litigation records from the Court of Friesland from 1716, 1718 and 1720. Network analysis was used to examine whether certain sources were more dominant (i.e. more central) in the network than others and to explore the relationship between the references. The lawyers in the three cases from the Court of Friesland appear to have used some references in common when arguing whether security rights (i.e. mortgages) included a right of pursuit and whether the auctioneer could recover the object if the buyer failed to pay.


Introduction
Civil court records are interesting for the legal historian, specifically from a perspective of law and its development over the centuries. The references to Roman and customary law that these records show demonstrate, for example, how the reception of Roman law functioned in legal practice in Friesland in the early modern period. Such cases are commonly analysed by intensive reading and note-taking and by identifying commonalities, differences and relationships between documents or elements of documents. This article focuses on the relationships, and more specifically on the allegations (citations, references), in a selected number of cases. We use these cases to explore whether human analysis can be improved, or at least complemented, by applying network analysis. The latter is a computer science method that allows for the mapping, measuring and visualising of relationships between individuals, groups and other types of information1. In this form of analysis, nodes are connected through edges, with the nodes being individuals, groups or information, and the edges being used to link the nodes. By treating court decisions as nodes, and linking the allegations in those cases to legislation and to other cases or scholarly work, network analysis can create a citation network that signals how information flows or has flown, and the extent to which certain nodes in the network are authoritative2. Network analysis has been used for several purposes in the legal field: for analysing criminal behaviour3 and terrorist networks4; for finding authoritative cases at courts such as the European Court of Human Rights (ECtHR)5, the Court of Justice of the European Union (cjeu)6 and the International Criminal Court, as well as in US7 and Canadian case law8 and Italian constitutional cases9; and for 5 Y. Lupu and E. Voeten, Precedent in international courts: a network analysis of case citations by the European Court of Human Rights, British Journal of Political Science, 42 (2012), p. 413-439. In this article, the ECtHR cites precedent based on the legal issues in the case instead of on the country of origin and states that it is more careful to embed judgments in its existing case law when it rules against a domestic court's decision than when it confirms the domestic court's decision. 6 M. Derlén  examining legal social networks10, networks of statutes and regulatory codes11 and patent citations12. Volkaert argues that network analysis can be both conceivable and useful in legal history. His recent article provides an overview of research in digital legal history using network analysis and focusing on case citation networks and what he refers to as 'digital-dogmatic legal history' . According to Volkaert, network analysis can complement dogmatic and contextual legal history, although qualitative juridical interpretations for understanding law remain necessary13. By applying network analysis to a selected number of cases of the Court of Friesland14, we further explore its potential for legal history research purposes15. We conducted the analysis using Gephi, an open source software for exploring and manipulating networks16. Regarding the data, we focused on three similar civil litigation records from the Court of Friesland from 1716, 1718 and 1720. These documents consisted of the litigation documents available for both parties, with most of these documents being available in their entirety17. The court records included the linen bag in which the files were submitted, the judgment, the plaintiff's claim, the defendant's response, various testimonies of witnesses and documents of a purely procedural nature such as a document granting leave to appeal or one inviting a witness to be present at the lawyer's office in Leeuwarden at 10 o'clock for an interrogation.
The focus of our examination was on the references (allegations) made by the lawyers in these cases to Roman and customary law as well as case law when arguing their case. Our research concentrated on the references and on establishing whether certain sources were more dominant (i.e. more central) in the network than others. Another question concerned the relationship between the references. References could be expected generally to concern Roman law, given that Friesland -in contrast to the other provinces in the Dutch Republic -chose Roman law to be applicable insofar as it was not set aside by the province's statutes18. Whether this expectation holds true will become apparent from the network visualisation.
The contents of the three court records and their coherence are briefly described in section 2. The legal question, the simple interrelationships and any more complex relations between the references in the three cases are also addressed, while the references to Roman law, customary law and case law are categorised. Section 3 visualises the network, as well as mapping and further exploring the references in the three similar litigations and discussing the relations visualised between the cases. The results, and broader remarks on network analyses in legal history, are discussed in section 4, followed by the conclusion in section 5.

2
Three similar Frisian court cases

Recovery from a third party19
For the purposes of the examination, three civil court records with similar cases were selected dealing with the same legal question20. Two of these cases could be joined by close reading the court records, while the third was found in a footnote in one of the other case studies21. The facts in these three cases were broadly similar in that all three involved an object being sold at an auction. In the first two cases, the object was a black mare, whereas in the third case it was a red spotted cow. All three objects sold were encumbered with a mortgage and all three buyers failed to pay the purchase price, even after being reminded. The auctioneer who had sold the object subsequently had to recover it from a third party. In all three cases, the latter refused to surrender the object, claiming that it had been sold to him in a legally valid way and that he was therefore an (unassailable) possessor in good faith. Each time, the corresponding legal question was whether the security right (i.e. the mortgage) was valid only if the object was still with the debtor. Or could it also be valid and invoked if the object had been sold to a third party? In other words: did the security right include a right of pursuit, and could the auctioneer recover the object from the possessor in good faith?
Although the facts in the three cases were largely the same, the conditions of the sales were not. In the first case (inventory number 10099: Abbema/ Douwes), the auctioneer had added the auction condition clausula constituti (bezit-accoort) and two sureties22, which the plaintiff initially failed to call 19 For a detailed explanation of the three cases, see H. upon. As a result, and as far as we can reconstruct23, the claim was rejected by the court on 18 February 1716. In the second case (inventory number 10415: Fenema/Heringa), the auction condition clausula constituti was not added, but a specific mortgage was taken out, and there was no surety. Here, it would appear, the security right included the right of pursuit. In the final judgment on 22 February 1718, therefore, the defendant was ordered to hand over the black mare or pay the purchase price to the plaintiff (the auctioneer). The conditions in the third case (inventory number 10705: Coehoorn/Symons) were the same as in the second case, with the exception of the mortgage. In this case, the mortgage was not a specific mortgage, but a general one, which implied the privilege of the right of eviction (voorrecht van uitwinning). The plaintiff should, therefore, have first sued the non-paying buyer, which he had neglected to do. The Court of Friesland thus rejected the claim on 6 February 1720.

2.2
Simple interrelationships Both the second and third case contains references to one of the preceding cases. In his reply, the plaintiff in the second case, Fenema, copied 26 articlesthe claim, the reply etc. were enumerated -verbatim from the first case. Indeed, he explicitly mentioned this first case at the end of his reply. In the third case, the plaintiff, Coehoorn, copied 11 articles verbatim from the second case24. These articles also happened to correspond with articles in the first case, although Coehoorn mentioned only the second case. The three lawyers would appear to have considered their cases to be similar and may have thought they could benefit from the earlier case(s). This interrelationship between the cases was found coincidentally, through close reading. New questions arise from this ( finding, including how did the lawyers acquire procedural documents from the other cases? Did they have access to them in some way? Was it common to refer verbatim to articles from other court cases? Interestingly, and even without going into the cases in depth, we can also identify another simple interrelationship: the auctioneers from the first two cases and the red spotted cow from the third case all came from Marrum, a village in the north of Friesland. Did the lawyers then work together? Did they exchange procedural documents? Linking more cases of a similar nature could provide new answers to questions about legal practice in the early modern period. Problematic in the study of court records is that it takes a long time to identify allegations in the text and to find relationships between them. One has to be both conscientious and lucky to recognise interrelationship between litigations. Moreover, based on the sample used for the present study it is hard to draw conclusions. And although the interrelationships mentioned in the three cases studies here were easy to recognise, recognition will be more difficult where the numbers of entities are higher.

Relations between references: Roman law, customary law and case law
Since the facts and the legal questions in the three cases under consideration are similar, it would not be surprising if the references to the sources also prove to be similar. And, indeed, this turned out to be true. The three cases contain a total of 55 references; of these, 30 were unique references, not including the two references to the other two cases studied here. There is thus an overlap: more than once, the lawyers used the same reference. It is difficult to identify patterns of references through close reading. Instead, therefore, we labelled the nodes (i.e. information entities) and analysed the relationships between them by using network analysis. Before applying this analysis, however, we first had to categorise the various sources to which the allegations referred, given that conclusions can only be drawn on well-defined entities.
For this article, we distinguished three categories of references, defined more or less by content, Roman law (15 times)25, customary law (10 times) and case law (5 times). The three cases are separated from these categories. We are aware that this categorisation is not exhaustive and can be more accurately expanded in future studies, focusing more on the defining of the information entities. Some shortcomings arising from the use of inadequate definitions are that the legal commentaries and works on Roman-Frisian law 25 Fragments referring to Roman law originated from the Institutes, the Digest, the Codex, the Novels and the authenticae.
were classified -for the sake of convenience -under customary law26. We also did not take the classification of the use of allegations into consideration, for example, are they copied or are they used by analogy? In addition, no distinction was made between data pertaining to the plaintiff and data pertaining to the defendant. The extent to which references were used at different stages in the legal proceedings was also neglected. However, the three categories chosen were considered sufficient for the purposes of this article, which sets out only to demonstrate the benefit of network analysis for this dataset.

Visualisation of relations
Visualising connected nodes leverages humans' perceptual abilities to discover patterns from data associated with nodes and edges. The network data in this article were acquired from external data sources27. The network is represented by nodes and edges: the nodes are the relational data such as sources (cases on the one hand, and Roman law, customary law and case law on the other hand), while the edges represent the citations between these nodes. On all occasions, the cases records are the sources because only these records contain references. The visualisation of the references in the three cases is presented in figure 1. The layout of the network is based on the Force Atlas 2 algorithm28, with the three cases illustrated in orange as set out below. As mentioned above, the first case has inventory number 10099 (1716), the second 10415 (1718) and the third 10705 (1720). The figure should be read chronologically, from left to right, with references to Roman law being shown in purple, references to customary law in green and those to case law in blue. The node size depends on the incoming references: the more incoming references, the larger the node size (or, in network analysis terms, the larger the in-degree value). The same applies in respect of the thickness of the arrows: the thicker the arrow, the more often the lawyer referred to the particular source. What can we generally deduce -despite the limited categories -from the network? The three nodes in the middle are the largest and therefore have the highest number of incoming references. All three cases refer to these fragments. The high (in-degree) centrality of these nodes suggests that these fragments are authoritative in this kind of legal issue. Two fragments from the Heedendaegse rechtsgeleertheyt by Huber are important, while one from the Codex is. The first case in time (10099) contains the most references, while the second case contains substantially fewer references. This second case refers to the first case, as the third case does to the second one. In its references, the third case is limited to the most important references, while the first two cases have some references in common, as shown at the top left. Both these cases refer to other case law in blue (16556). This case is from 20 December 1687. The second case also refers to other case law, specifically the judgment issued on 1 February 1676 (shown as 16545).
Roman law to be applicable. If customary law could be further specified within the category, for example, of Roman-Frisian law, the reception of Roman law could be calculated more precisely. The more accurate the definitions of information entities are, the more specific the information generated will be. However, qualitative juridical research will continue to be necessary.

Future of network analysis in legal history
For what other legal questions may network analysis be useful? As well as identifying meaningful relations between entities within a dataset, it could be helpful for combining similar sets of entities29. If other court archives in the Dutch Republic were to be unlocked, relations between these courts could also be revealed. Since each province was autonomous in its private law, litigation in each court is likely to have had its own character. References in litigation before the Court of Holland, Zeeland and West-Friesland sometimes contain references to Frisian case law30. But was this common? Did lawyers from the province of Holland refer to Frisian jurists? If so, why and when? Can we find lawyers practising in more than one court? What can we discover about the reception of Roman law in other provinces' courts? What differences in this respect can be discovered in the judgements of different courts? What developments can be identified based on these judgments? Are these differences expressions of identity? And what if we were to combine the set of entities with case law and consultation books31? Not to mention questions pertaining to other disciplines such as history, sociology and economics. Today, entire archives of centuries-old material are available digitally and have been made searchable. Although such initiatives have commonly been undertaken in (digital) history departments, initiatives in the legal domain are under development. Fortunately, important initiatives are taken to change this situation. In, for instance, a pilot study, focusing on legal history, at the The resultant data will be from 2273 civil cases from the period 1716-1730 (all documents in manuscript)32. This will be the first time that such a large amount of historical material from legal practice will be available. A challenge resulting from the increased availability of data is that identifying relationships between documents or parts of documents become unmanageable if only human analysis is applied. While a citation analysis, as conducted in this article, may be feasible with three cases and around 50 references, it would be impossible to conduct such an analysis with 100, 500 or 2,000 cases (nodes) or references (edges). Computational methods can assist in analysing historical data by, for instance, automatically recognising references in the text. This can then result in more data becoming available metadata; that is, data about the data (e.g. citations). The availability of more digitalised documents in the field of legal history will also allow leveraging of computational methods and techniques for analysing the documents, with network analysis then being used, for instance, to explore relationships between large numbers of cases and between references to and from those cases. However, this type of research requires new (digital) skills and implies a transformation, at least in part, of academic research in the sense that the very nature and scope of historical research will change33. Moreover, although promising, network analysis has its limitations. For example, particularly in large networks it becomes possible to group nodes (e.g. decisions, cases, arguments, sources) together based on their position in the network. Several algorithms exist for conducting such community detection, yet there is a lack of clarity as to precisely which algorithm is the most optimal one in which type of network. The same holds for determining the centrality of nodes. In-degree (incoming references) and out-degree (outgoing references) are intuitive measures, but a wide variety of additional measures exist, for instance algorithms that take into consideration the centrality of the nodes where references come from or go to (e.g. hits, PageRank) or that give more weight to references from or to nodes that are younger or published more recently compared to other nodes. 32 Tresoar, Leeuwarden, Archive of Court of Friesland (acf), inv. no. 10089-12361 (ca. 48 metres). As each metre of archives contains ca. 4,080 images, 48 metres will contain ca. 195,840 images. These records comprise only part of the total of 196 metres of archives: 17,030 inventory numbers (acf, inv. no. 7599-16470

Conclusion
This article discusses three similar cases, in which the legal question at stake was whether security rights (i.e. mortgages) included a right of pursuit. Could the auctioneer recover the object if the buyer failed to pay? The answer to this question depended on the contractual conditions. The lawyers in the three cases from the Court of Friesland appear to have used some of the same references. Network analysis of these references was used to visualise the relationships between these three cases, with the most important references also being made visible. A network can also be customised to suit specific purposes: the more entities (big data) that are defined, the more information that can be generated. What will the future of network analysis in legal history bring us? Once the 48 metres of civil court records have been labelled, network analysis can be used to expose patterns and trends that cannot be observed by the naked eye, including, for example, relations between legal problems, geographical locations, lawyers and clients. It can then be further leveraged, particularly if references can be detected automatically, to detect relationships between cases, references and actors that cannot be detected by means of human analysis. By comparing legal practice in civil court records in various provinces, or even internationally, network analysis will also make it possible to explore the mode of operation in these courts. This type of research in legal history will then generate new information that will add knowledge to historical legal practice, as well as uncovering information about daily life in early modern history and possibly also leading to a new understanding of the identities of the various provinces.