When a plaintif or a defendant finds themselves engaged in a legal case, they often turn to legal practitioners for getting perspective on their legal standing and how the argument and conflicts play a part in the court's decision making. Since much of the past legal proceedings are accessible online, many individuals seeking legal counsel use a search engine to obtain some knowledge of related historical cases on their own. In this research, we present a conceptual framework based generation of legal case perspectives for further case planning. For previous legal proceedings, we construct conceptual frameworks led by a surrogate template generic to all court cases. We generate concept classifications and indications using these conceptual frameworks, used later in the text search to extract relevant text spans to generate numerical analyses for getting perspective on the historical data. We provide users with a case characterization system, in which users select case elements that apply to their situation. We construct the numerical perspective evaluation based on case characterization by parsing the text from historical cases. We explain our system using two case studies: Divorce and Parental alienation cases demonstrating the efectiveness of the system in case planning.
cedure for strategic case planning to produce legal case Nowadays, a large volume of legal data is available online perspectives. Both the clients or the parties contesting in the form of court cases from specific courts, articles, the legal case and the legal experts can use the system acts, and other secondary materials. When individuals to gain information from similar prior proceedings, thus encounter a legal situation, they look for these points helping them prepare their next steps in the case. of information to get a legal interpretation of their case. For generating the legal perspectives from past cases, However, because of the formal and specialized vocab- we present a generalized surrogate template applicable in ulary used in legal documents, they may find it hard to most of the court’s cases. The surrogate template drives understand it. As a result, people seek legal counsel from the creation of conceptual frameworks for the data. We practitioners to anticipate their position in the case and create concept classifications and indications from the what factors impact the decision-making of the trial. conceptual framework. For filtering the past cases that
Likewise, legal professionals may also choose to ana- are similar to the user’s situation, we provide a case charlyze historical data to gain insight into the interests of acterization system. Using text search and parsing on the their customers. Nevertheless, it is a tedious job to look filtered data, aided using concept classifications and infor client-specific cases and understand the significant dications, we generate numerical perspective evaluation insights from them. that enables planning the future case-related activities.
Current research in this area explores the use of natural Two of our specific contributions are:
language processing on legal texts to assist legal profes- • In a way that is readily accessible by both legal
sionals in various ways. Case summaries [
Following is the organization of this paper. In the Section 2, we explain the method. In Section 3, we explicate our approach using two legal spheres, namely divorce and parental alienation. In Section 4 , we address the findings. We discuss the existing research and summarize the paper in Section 5 and 6, respectively.
We note that all legal proceedings consist mainly of six el- which correspond to the concepts. However, there may ements. Four non-derivative elements are the parties con- be phrases that are representative of such concepts. We cerned, previous decisions (on an ongoing case(s)), the process the case files to classify those. We apply standard parties’ facts, and the parties’ appeals. The two deriva- normalization to the text. Countries around the world tive elements are the decision of the court and the reasons document their cases in various ways. They may contain for the decision. The derivative elements are the outcome diferent section names or no sections at all, still commuof non-derivative elements in the case. Using these ele- nicating all the information mentioned in the surrogate ments, we build a surrogate template to represent legal template. In the cases under consideration, we map the cases, as shown in Figure 2. particular set of sections into six elements specified in We map the text into these six surrogate template ele- the surrogate template. If there are no sections in the text,
one needs to map the text to various surrogate template elements manually.
We collate the element-wise text from all the files and cluster the sentences. We vectorize the text using TF-IDF vectors and use Principal Component Analysis (PCA)1 for dimension reduction. We apply K-means Clustering2 to the resultant embedded text. We keep the number of clusters, as a variable that can be adjusted based on the results one is getting. We get relevant clustered sentences for = 10. From each group, we show 20 sentences at random. The expert goes through the sentences and identify any phrases that represent a concept, and add it to the indication dictionary.
We use the indication dictionary later for search and parsing operations for getting the relevant text spans to generate the numerical perspective output.
System interface shows the question statements and options from the case characterization system. Based on the options selected by the user, the indication dictionary provides the terms and phrases to search the text. We calculate the numerical statistics of count and percentages 2.3. Case Characterization System from the text spans where values of indication dictionary appears. We display text spans and statistical results as The designing of the case characterization system re- the output. quires the use of the concept classification dictionary . This system creates question statements and options for the users to characterize their case. We create rule-based 3. Case Studies question statements from the non-derivative elements We describe two case studies: parental alienation (PA) and divorce proceedings. Parental alienation often re1Principal component analysis in sklearn https://scikit-learn. ferred to as PA, is a condition in which, one parent emorg/stable/modules/generated/sklearn.decomposition.PCA.html ploys techniques such as manipulating, isolating, or congene2rKat-eMde/saknlsearcnl u.csltuesrtienrg.KMhtetapns:s/./hstcmiklit-learn.org/stable/modules/ ditioning to distance a child from the other parent [18]. of the surrogate template. We add “Select the most appropriate" before the non-derivative element followed by “categories for your situation". We parse the concept classification dictionary to obtain the corresponding options. Since these questions are case category-specific, we add one more question statement before all of them: “Select the case category" for filtering the legal domain.
The options for this statement are all the legal domains available in the dataset.
Divorce refers to the separation of husband and wife. The and then the expert creates the element-wise conceptual rationale behind choosing these civil spheres is that the frameworks. psychological and mental health of the people involved Due to limited space availability, we show a succinct is significantly influenced by both parental alienation representation of conceptual constructs for divorce and and divorce cases [19]. In helping the parties to under- parental alienation cases in Figures 3 and 4 respectively. stand the dynamics of such situations, a framework of The hexagonal structures are part of the surrogate temlegal perspectives may be helpful, for instance, to a poor plate that informs the conceptual framework modelling. married spouse or an estranged parent. The classifications of the groups that appear in each sur
Statistics3 show that Europe is the top continent with rogate template element are the rectangular constructs in the highest divorce rates in the world. Divorce is one of the conceptual frameworks. Solid lines in the figures rethe major causes of parental alienation. Due to the easy lfect how the surrogate template elements relate to each availability of data, we use legal cases from Dutch civil other. We also show dotted lines that explicate how the court. The Netherlands4 is in the top 10 countries with classifications relate to each other. We only show two the highest divorce rates in Europe. sample classicfiation concept relations due to space re
For PA5 and divorce6 proceedings, our case datasets strictions. The figures also show the concept classification contain 109 and 102 case files, respectively. dictionary, where the keys are the hexagonal structures
Using Google Translate API7, we change the language (surrogate template elements), and the values are the rectof legal documents from Dutch to English. The translated angular elements (the classifications in each element). data of the Dutch civil court already contains sections Composition of PA and Divorce Cases The cases mappable to the surrogate template elements. If the data of parental alienation apply specifically to child custody has varying section names, then one can manually cre- cases. Multiple stakeholders, such as parents, relatives, ate a section-mapping dictionary to map the sections. foster care, and institutions (such as youth care centres) If there are no sections, one needs to add the relevant may start the cases. The key subjects of the dispute are section names to the text. We collate the resultant text, the residence of the child, provisions for contact among the parties and the child, custody of the child, supervision of child with child care services, and the expenses of child development and legal proceedings.
The cases of divorce between men and women are contested mainly with the participation of other stakeholders.
The cases primarily concern marriage disputes, costs and fees to be borne by the spouses, including costs of child care, housing expenses, costs of subsistence and litigation costs, distribution of marital assets, arrangements 3Divorce demography https://en.wikipedia.org/wiki/Divorce_ demography
4Divorce rates in Europe in 2017, by country https: //ec.europa.eu/eurostat/statistics-explained/index.php/Marriage_ and_divorce_statistics
5Sample file available at https://uitspraken.rechtspraak.nl/ inziendocument?id=ECLI:NL:GHAMS:2019:44, more files available with other nomenclatures.
6Divorce cases https://jure.nl/echtscheiding 7Google Translate API https:/translate.google.com/
Text Processing and Case Characterization The expert creates the concept classification dictionary using the conceptual framework. We process the case texts and cluster the sentences to aid the expert in identifying additional references of the concepts to generate the indication dictionary. Figure 5 shows the sample keys and values for a few concepts for divorce and PA cases.
As described in Section 2.3, we create the question statements for the case characterization of the user shown in dark green color in Figure 6.
We parse the concept classification dictionary for providing the options for each question statements. For creating the numerical analysis, we parse the text using the indication dictionary values based on the options chosen by the user, which are the keys of the indication dictionary.
In our case, they are Divorce and Parental Alienation. 4.2. Parties Based on the user’s choice, further options are based on the legal domain-specific concept classifications as Input-Output The second question statement presented shown in red and blue colour in Figure 6 for divorce and to the user is Select the most appropriate parties catePA respectively. gories for your situation. The user is shown the options
The output returns numerical information on the in- for all the parties involved according to the legal domain volved parties, and the text spans from which the nu- chosen in the first step. We present the outputs for the merical results are calculated. The outcome indicates situation when man, woman and council are selected for the number of parties that usually compete in the se- divorce cases, and mother, father and council are chosen lected court cases and the top group of parties fighting for PA cases. together. Figure 7(a),(c) shows the party count and Figure The choice of the parties yields the text spans and 7(b),(d) shows top group of parties for divorce and PA numerical results on previous decisions and the parties cases respectively. associated with each category of the previous decision.
Perspective and Case planning Parties’ data dis- Figure 8(a), 8(c) shows percentage distribution of catecloses general stakeholders in a civil sphere. It also in- gories and Figure 8(b), 8(d) shows a stacked graph of cludes details on the groups which mostly appear to- counts of parties in the previous decision for divorce and gether, which helps the non-legal user explore who may PA cases. get involved in their cases that the user might not have Perspective and Case planning The results for direalized previously. For example, man and woman are the vorce cases show that divorce, previous appeals, the restop parties in divorce cases. Nevertheless, council, man, idence of the child, and the arrangements between the and woman are the top-most occurring group in divorce parties are the top previous decision categories. Similarly, cases. The user also learns that there may be some cases in PA cases, they are child placement with the institution, involving specific organizations, notaries, curators. arrangement between the parties, and previous requests.
Similarly, the father and mother are the top fighting The results also convey the association of the parties in parties for child custody in PA cases. The council and various categories. Divorce results show that man and the institution are part of the top group involved in the woman plays a vital role in most of the previous decision PA cases alongside father and mother. Also, the user categories. It shows that man is more engaged in sperecognizes that there may be other parties that may be cific categories, such as costs, income, and arrangement. a part of the case, such as foster parents, relatives, and Other categories such as agreement and divorce have an guardians. equal engagement of both man and woman.
The legal user also gets this information readily avail- Similarly, the PA results show major involvement of able to them. The legal professional might like to intro- mother, father, council, and institution in various cateduce or use one of those parties in the current case to get gories. The institution and council are more likely to be maximum benefit in their appeal. involved in child placement decisions and previous requests. Father and mother are concerned with topics like arrangement, custody, child placement, and residence.
For a non-legal user, these results show insights on the topics where decisions get pending, or some actions
most appropriate previous decisions categories for your situation.
We present the results for divorce cases where the user chooses to divorce, parental plan, costs, and income categories and for PA cases where the user selects residence and arrangement categories.
The selection of previous decisions leads to numerical results and text spans of the facts of the cases. We show the percentage distribution of fact categories and the Figure 8: Party results: (a) previous decision categories in parties involved in facts for divorce cases in Figure 9(b) divorce cases, (b) parties involved in previous decision for di- and 9(a) and analogously for PA cases in 9(c) and 9(d). vorce cases, (c) previous decision categories in PA cases, (d) Perspective and Case planning Divorce outcomes parties involved in previous decision for PA cases. show that facts related to divorce, residence, child, and nationality are the most significant. The facts of divorce are relevant to both the man and the woman. But, mostly can be requested from the parties. For legal profession- the facts related to the income, costs, child, and residence als, perspective on previous decisions can be extremely of the child are related to the man. Likewise, for PA cases, beneficial. They can plan the case around the relevant supervision, residence, and custody-related facts are the topics to get the court’s decision in a particular way, for most significant. The mother plays a vital role in the example, if the attorney needs time to search more infor- arrangement, custody, and residence of the child. mation or evidence in a particular circumstance, appeals Non-legal users discover the crucial facts that may help around certain subjects may stall the case till the next in their situation. By presenting persuasive evidence and hearing. facts on specific topics concerning certain parties, the relevance of the facts revealed through the numerical fending party in one of the cases. Similarly, the father and results can assist the legal professionals in designing the mother are appealing and defending parties, where their case. the mother appeals the most. Council is mostly defending the cases followed by the mother and the father. Most appealed categories in divorce are costs, a previous decision, 4.4. Facts property, and residence. In the case of PA, a previous deInput-Output The fourth question statement is Select cision, supervision, and custody are the most significant the most appropriate facts categories for your situation. appeals.
We present the results for divorce cases where the This perspective informs the non-legal users about user chooses the divorce, costs, and income categories the critical appeal topics and the parties who appeal and and for PA cases where the user selects residence and defend. Non-legal users may prepare their appeal or arrangement categories. defence by comprehending how the appeals were made
The selected fact categories show the text spans and in the previous cases and how can they be presented in the results related to appeal categories and statistics on the current case. the appealing and defending parties. Figure 10(a) and 10(b) shows the appeal-defense count of the parties and 4.5. Appeals the percentage distribution of appeal categories for divorce cases. Figure 10(d) and 10(c) shows the same results Input-Output The last question statement is Select the for PA cases. most appropriate appeals categories for your situation.
Perspective and Case planning Divorce results We show the numerical results for divorce cases when the show that the man and the woman are the parties who chosen options are property, costs, a previous decision, appeal for a particular dispute, and the man appeals more and for PA cases when the chosen options are custody than the woman in this situation. The man is also a de- and supervision.
This selection displays the final results and the text present a systematic review of legal ontology literature, spans of the decision and the reasons for the decision. easy interaction, and creation with legal ontologies, inThe reasons categories and the decision for divorce cases formation extraction to represent them as ontologies. are shown in Figure 11(a) and 11(b) . The same results Conversely, we present a conceptual framework of the for PA cases are shown in Figure 11(d) and Figure 11(c). legal database made from a generalized surrogate tem
Perspective and Case planning Divorce cases show plate that is used to produce legal perspectives for the
half of the cases ratified and half of them annulled. The end-user based on their situation.
most relevant reasons for the decision are judgment, Legal Text Summarization Various NLP and machine
costs, and property. In PA cases, half of the cases are learning techniques [
Decisions reveal the probable outcome of the case are decision or case-specific. Unlike our method that is
acbased on the relevant reasons categories for the non- cessible for both legal and non-legal consumers, reading
legal users. This perspective aids the legal user to plan the summaries can still entail some legal expertise.
their case. The prosecutor may plan to show substantial Legal Information Extraction Many researchers have
facts and evidence relating to the most relevant factors introduced various techniques for diferent types of
leshown in the reasons that may support their argument gal information extraction like legal entity extraction
and disprove the appeal of the opponent. from legal documents [
In terms of the use of mathematical modelling-based knowledge extraction provided by our methodology, we diferentiate our approach from the works mentioned above. Also, our approach focuses on providing perspectives for users to support them in further case planning.
Following are the limitations of our study and the way forward: • We generate the results using separate sections under each question statement. For this, we manually map the text into the surrogate template elements. Since the data may contain no section or sections in diferent formats or with variable names, we plan to introduce a classifier to divide the case text into specified surrogate template elements. • We rely on the manual intervention of legal experts for the conceptual framework development (supported by [17] to aid the creation). In the future, we consider using word embeddings, relation matching, and ranking to strengthen the expert’s recommendations. • The output shows the text spans from which the numerical results are obtained. For a large number of files and huge text spans, it becomes tedious for the user to read all the text. In the future, we plan to provide text recommendations to the user by comprehending and categorizing meaningful text using structured prediction models, which requires a huge amount of annotated data. It is part of our ongoing work.
Legal Ontologies and Applications various
researchers widely research legal ontology creation and
its applications [
Historical court proceedings may provide users with perspectives on the parties that are often engaged in a case, pertinent facts and past rulings, what appeals are being made and the parties that are appealing and defending, and the key reasons for diferent decisions. These insights will help users prepare their case carefully in order to achieve the optimal result in the current case. We present a conceptual framework guided legal case perspectives for divorce and parental alienation cases, proving the usefulness of the system. In the future, we plan to work on the manual method automation and fine-grained information presentation in the production.