® InfoJur.ccj.ufsc.br
Instituut voor Taal- en Kennistechnologie
Institute for Language Technology and Artificial Intelligence
Faculty of Law / Institute for Language Technology and Artificial Intelligence (ITK)
PoBox 90153
5000 LE TILBURG
The Netherlands
This article is about LEDA, a legislative drafting support system using hypertext technology. In the development of legal information systems, hypertext technology is getting more and more recognised for its powerfull interface properties. In the LEDA system we have extended the use of hypertext for the visual representation of legislative knowledge. Starting off with considering the nature of legislative reasoning we will show how hypertext technology combined with knowledge based templates can be used to create an information environment for integral legislative drafting support. We further present a method for modelling hypertext networks and dynamic creation of hyperlinks using a frames formalism.
Over the past ten years the Dutch government has - due to some serious problems regarding the effectiveness of legislation - become increasingly concerned with legislative quality. To improve the overall quality of legislation, different policies were pursued and enacted [Legislation in perspective, 1991]. One of the main results of these governmental efforts was the adoption of a general legislative policy, which consists of a set of measures aimed at the lasting improvement of legislative quality by setting quality criteria [Eijlander, 1993]. All of these legislative quality requirements are Ä together with legislative technique-requirements Ä included in the Recommendations for Regulations which the Dutch Prime Minister issued in 1992 as an instrument to further implement the goals of the general legislative policy [Recommendations, 1993].
The Recommendations for Regulations consist of 346 directives and guidelines Ä addressing the legislative draftsmen within the ministerial departments Ä regarding important drafting issues and activities. Aside from legislative technique issues, like terminology and model clauses, they also deal with policy aspects, methodological issues, procedures, structural design etc. Although they closely resemble ordinary legal rules, they are of a different nature. They are not always generally binding, like legal rules, but directives that can, in certain cases, be ignored if there is a good reason to do so. They constitute a mix of legal (constitutional) rules and guidelines concerning `best practices and solutions', derived from legislative experience. Besides legal rules, best practices, and legislative quality criteria, a large amount of quality safeguards are incorporated throughout the 346 Recommendations. The Recommendations therefore can be considered a voluminous `Draftman's handbook' dealing with every important activity within the drafting process (see the design-step- model in section 2). Related to the activities in the drafting process, the Recommendations can be categorized into the following groups:
The text of the Recommendations is not organized along the chronological and methodological lines of the drafting process, but rather thematically in the order of diminishing abstraction. This circumstance makes it quite difficult for legislators (even experienced ones) to find their way through the Recommendations during the drafting stage. An information system, it was felt, could be the way out of these problems. This meant the start of the LEDA-project.
The main goal in the LEDA-project was to make the information of the Recommendations accessible in concordance with the information-need during the different stages of the drafting process. A secondary goal was to make the information, referred to in the Recommendations themselves (secondary information), available to the users. Many recommendations, as it happens, do not prescribe what the solution has to be in a certain situation --- as is often the case with ordinary legal rules --- but rather prescribe which activity should be undertaken at a certain moment, and what kind of information is needed to be able to perform the prescribed activity. The third goal of the LEDA-project was to offer knowledge-based drafting-support on the basis of the legislative knowledge within the Recommendations, pursuant to the knowledge-based access of the information from the Recommendations.
In 1993 the project resulted in the prototype LEDA-system, which is currently used by a group of legislative draftsmen within the Ministry of Justice.
Drafting regulations is not just a matter of putting down legislative or policy choices into words, but involves a decision-making process in which many substantial choices regarding content, structure, structural elements and --- eventually --- phrasing and wording of a draft have to be made. Quite frequently legislative drafting even means that policy decisions have to be made or reviewed. In this, legislative drafting differs substantively from legal reasoning. Legislative reasoning is only partly dependent on legal problem solving, legal knowledge and legal reasoning. In comparison with other forms of legal reasoning or problem solving (like application of the law), legislative problem solving, i.e. the decision-making process aimed at the enactment of legislation, is much more dependent on world knowledge (common sense), and it equally involves, throughout the different stages, substantial political, socio-economical reasoning [Snellen, 1987, Habermas, 1992]. Furthermore, the legislative process does not primarily result in legally (in)valid conclusions, but rather in 'relatively appropriate' solutions, or in convincing arguments [Hotz, 1984]. Whether a bill is an appropriate answer to a legislative problem does only partly depend on its legal quality, and, vice versa, the correct application of legal requirements does not automatically procure good or appropriate bills [Voermans et al., 1992]. In fact the legislative drafting process is an open-natured decision-process in which a lot of different natured choices have to be made.
While making these choices a lot of requirements have to be met. These requirements are not only of a homogeneous nature, comprising legal standards (e.g. constitutional standards) and aspects of legislative policy and technique, but also of a heterogeneous nature resulting from various conditions related to particular subject matter, or from existing policies regarding the field of the projected draft. The drafting process is a complex decision-making process which requires great skill and knowledge. In the Netherlands most of the legislative drafting is therefore carried out by specialists within the ministerial departments. To ensure the quality of their drafts, these legislative specialists --- in most cases --- approach the drafting process methodically. Although these approaches vary between the different departments, some general characteristics of these approaches to legislative design can be discerned. Generally speaking, these approaches consist of the following (iterative and interdependent) steps:
This model of the legislative design process (the design-step-model) does not always concur with the actual designing procedure. According to the nature of a specific project, sometimes only a few steps within this process are deemed necessary. Sometimes steps in this iterative cycle are repeated. Analytically speaking, however, this process model is empirically and prescriptively substantiated. This analysis model also constitutes the pretext and the knowledge-backbone of LEDA.
In contrast to judges or administrative bodies that have to apply legal rules in individual cases (and thereby are interested in finding answers to legal questions) draftsmen are interested in efficiently and effectively finding arguments to support their case (i.e. a draft). They are not interested in representing the legal rule in any kind of logic, because when a legal rule has to be drafted it is not possible to fix the correct reasoning or correct deontic expression within that rule in advance, as it is possible to represent the reasoning already laid down in a draft rule for purposes of logical evaluation. As mentioned above, in the drafting stage, however, this logical evaluation is not in great demand. What the legislator wants is a political or socio-economical evaluation of his draft. For this a model or a method for the way in which a legislative draftsman can or has to find the (relevant) arguments supporting his case, is needed.
The knowledge about the drafting process (concerning legislative issues like procedural or technical knowledge) extracted from the Recommendations could be used to guide the draftsman through the body of information which contains the different kinds of arguments needed. The first step therefore is not to represent the (legal) rules held within the Recommendations logically, but rather to conceptualise the legal knowledge held within the legislation process and Recommendations itself. Since the analysis of the drafting process and the Recommendations proved that drafting activities rely strongly on information organised in a network, we found that hypertext technology was a suitable candidate for the technical implementation.
From a functional point of view the hypertext technology aims to enable users to make their way through a body of complex information in a manner that facilitates its ready appreciation or visualisation [Mital et al., 1992]. From a more conceptual point of view, hypertext technology provides the means for non-linear text organisation in computers by associating (parts of) windows on the screen with objects in the database and providing links between those objects both graphically (as labeled tokens) and in the database (pointers) [Conklin, 1987].
To make this possible hypertext networks possess nodes and links governing the relationships between the various nodes. Links and nodes can have a variety of types and properties. Nodes can, for instance, consist of (or better: correspond with) database objects which 'contain' chunks of textual information, but they can also contain (a piece) of a knowledge-based template, which contains hypertext links in its turn. Links can connect nodes in different ways. To establish this connection they can consist of simple or quite elaborate (knowledge-based) procedures. There are two methods for explicitly linking two points in a hypertext network: by reference and by organisation. The referential method supports non-hierarchical (for instance: associative) linking of nodes. The organisational method on the other hand explicitly creates hierarchical connections by connecting a parent node with its children, thus establishing a strict tree subgraph within a hypertext network graph.
A classification of links orthagonal to the reference-organisation partition is the one in static and dynamic links. With this we make the distinction between links that are predefined and cannot be changed, and links extending the network that can be derived based on new information which comes available either by input from the user or the triggering of knowledge processing rules.To extract the legislative knowledge from the Recommendations and to formally design the hypertext network we used an analysis frame, based on a quite traditional model of the different components or elements of a norm [Ruiter, 1987]. Each separate recommendation was analyzed with the following terms derived from the norm-element model:
Recommendation (norm) object:
Recommendation (norm) condition:
Recommendation (norm) operator:
Recommendations (norm) subject:
This model in its most abstract form can be instantiated with recommendations of all the different types mentioned in the introduction. For the design of the main hypertext support facilities based on knowledge from specific groups of Recommendations, the norm-frame model is extended and specified [Voermans et al., 1993a]. In section 4. we will introduce specific analysis frames based on this model for:
Before the technical discussion of these specific hypertext functionalities we will first briefly describe the architecture and the functionalities of the LEDA system.
LEDA is a prototype Legislative Design and Advisory system designed to offer access to the Recommendations (and secondary information) in a methodical way, concurrent with the stages of the drafting process and through this, offer knowledge-based support for the drafting activities of legislators regulated in the Recommendations. To this end LEDA contains four major (integrated) functionalities, namely:
These functionalities are integrated throughout the system and can best be discussed by a description of the functionalities of the system's modular components. LEDA consists of two major modules:
The Preparatory Module in LEDA was set up to offer knowledge-based access to the Recommendations concerning substantive, methodological and structural design issues, in a way consistent with the chronology of events in the drafting stage (see for this chronology the design-step-model in section 2.).
The Preparatory Module (PM) combines the functionalities of a hypertextsystem with a knowledge-based (KB-) template system. The hypertext-based PM of LEDA permits the user not only to draft a preparatory document (e.g. a policy memorandum) but also supports the creation of a skeletal form for a KB-template, to be used for the actual structural design and formulation of a draft (Basic Design Screen). To this end the Preparatoy Module guides the user through a hypertextnetwork of semantic hierarchical and referential links. To offer guidance, the hypertextnetwork of the PM is divided into different levels, corresponding with the different methodological steps of the design-step-model derived from the Recommendations. The levels in their turn serve as a checklist, expressing important points of attention regarding methodological aspects and the structural design of a draft. A look at the systems' interface architecture (which closely resembles the functional architecture) may illustrate these features:
Fig 1: LEDA: Interface-architecture (functional modules and components)
The Preparatory Model consists of various methodological and consecutive levels (dotted lines on the left hand side). These methodological levels are referentially linked with level information (box at the upper right hand side). The level information component consists of (access to) the relevant Recommendations, access to relevant secondary information (as referred to by the relevant recommendations) and a graphic template-scheme for user-analysis of certain options. Level information changes according to the level which is active (i.e. the level in which the user is working).
The methodological levels themselves consist of fields containing information (about what is to be done within a certain level) and knowledge-based templates. The level-template-documents which mainly serve to insert (or draft) text, also support the identification of important sub-items and the choice between options. Both on the basis of the choice of the user and automatic analysis of text-input in the template, the system makes inferences regarding the arrangement of levels further down the network's path (e.g. the arrangement of the levels in the Basic Design Screen). From the point of view of the user, the levels form an interactive word-processor which provides methodological guidance and provides relevant (semantically interlinked) information.
The user may progress randomly through the level-structured hypertextnetwork. This fundamental openness of the system is necessary as the user-legislator is always free Ä when drafting a legislative text whithout the use of the system Ä to deviate from the Recommendations themselves whenever there is a good reason (as described in the introduction). To accommodate reluctant users, there is even a possibility to shut down the levels altogether. What remains is a word-processor linked to information in a single default-information level explaining the methodological approach of the Recommendations, and providing (links to) the relevant recommendations and secondary information.
To prevent getting lost in the hypertextnetwork, user-guidance is provided by the levels themselves, together with easy backtracking procedures and a step tracer, which consist of a major and minor active compass which visibly record the path hitherto followed in the network. On top of this, the PM is provided with a General Information-component to offer non-hypertextual access to various internal or external databases. Users can retrieve text from these databases while working in the different levels. The text in the internal databases, however, is hypertextually linked.
The Basic Design Screen Module (BDS) is developed and structured in a way very similar to the Preparatory Module. Like the PM it consists of a level structure, linked with level information. The levels (see the dotted line in the BDS-module of figure 1) contain templates mainly consisting of free-text fields, which allow system supported insertions (e.g. of model clauses or examples). The templates within the levels of the BDS however do not express points of attention with regard to the preparation and structural design but important phrasing, terminology and terminology-related (substantive) issues regarding the structural elements of a draft. The arrangement of the levels in the BDS is both based on knowledge (gained from the Recommendations) and knowledge-based inferences made by the PM module. The BDS itself can be regarded as one large knowledge-based template which is shaped and directed by the PM. The BDS represents the preferred structure of a draft, modelled to the needs of the user.
Like the PM the BDS has a very open structure: the user may progress randomly, do away with the levels altogether and receive default-information, and delete or add certain levels. The user-guidance function is similar to the one in the PM. The BDS has, however, one distinctly different feature compared to the PM.
Apart from the statical support of the referentially and statically linked level information (figure 1) the BDS offers a dynamic form of disclosing the Recommendations. The dynamic information support uses a parser (operating with a form of the linguistic conceptual dependency method [Schank et al., 1981]) to dynamically create hyperlinks from the drafted text to relevant recommendations. This facility accomodates the interest that the user may have after finishing the drafting of his text, whether he has overlooked relevant recommendations. In section 5. we will elaborate on the subject of the dynamic linking process. We will introduce objects to represent the subjects of the Recommendations and indicators to relate them to the draft text.
By considering the nature of legislative reasoning we have shown hypertext to be cut out for legal drafting support. In section 3. we described the integrated information support environment that the LEDA system offers by the use of hypertext. In this we have gone further than the current state of the art where hypertext technology was primarily used as an interface tool. By organising the knowledge of the legislative drafting process in a hypertext network we have used hypertext as a form of visual knowledge representation. With this, it is very useful as an intermediate language for conceptualisation of legislative knowledge (incorporated in the Recommendations).
We feel that hypertext is a useful knowledge representation technique, where the use of a network of typed nodes and links gives us the power of semantic networks and the possibility to represent procedural knowledge in the execution of links (following the links) as in frames representations, adding the possibility of dynamically creating links (based on knowledge present in the network or interactively entered information). This however calls for a formal model of hypertext. There has been little research in formalising hypertext for knowledge representation purposes. Recent publications of formal hypertext models either concern the use of hypertext as presentation tool [Halasz et al., 1994] or work in information retrieval [Bruza, 1993]. Closest to our ideas comes the work of [Neubert et al., 1992a,b] in using hypertext as intermediate language in knowledge acquisition and conceptual modelling. Research on the topic of hypertext as knowledge representation technique is ongoing. In section 5. of this paper we will illustrate the use of a particular frame-formalism to represent the dynamic links.
In the Preparatory Module of LEDA the different preparatory activities, regulated in the Recommendations are represented in a methodological way. We have pointed out already that the Recommendations are not arranged methodologically, but thematically. In order to be able to offer methodological guidance and assistance in LEDA, we first had to distil the methodologically important issues and activities from the Recommendations, and assess their interdependencies. To discover the methodologically important elements, we used an analysis-frame, based on the norm-element model [Ruiter, 1987] mentioned in section 2.
The analysis frame is shown in table 1. It enabled us to distinguish separate activities in the preparatory proces of legal drafting and to analyze the relations between them.
Recommendations_Object
[
/* Recommendation Object (Activity) */
activity type : {answer, choice, scheme};
activity trigger : activity_frame.status;
required information input : set_of_string;
information output : set_of_string;
/* Other Norm-elements*/
Rec. condition : logical_expression;
Rec. operator :{obligation, permission,
prohibition, command};
Rec. subject: string;
]
logical_expression : string;
Table 1: Recommendations analysis frame
On the basis of this analysis we were not only able to distil the preparatory legislative activities and their interdependencies from the Recommendations, but we were also able to construct a hierarchical frames-representation of the different drafting activities themselves, and their mutual relations. The latter operation resulted in a model which closely resembled the design-step-model discussed in section 2, consisting of seven major consecutive design steps. Within the design steps of the model, several interrelated substantive (subordinate) activities, issues and questions, regarding the preparation of a draft and the draft structure, are represented in their turn. In this way the analysis resulted in a methodological transposition of the Recommendations into a design-step-model.
An obvious advantage of the frames representation in the model was, that we were able to assess the information-basis of the different activities formulated in the model. This resulted in the conclusion that, although many activities were information-based, they relied on formally representable (e.g. legal) knowledge only to a very small extent (see also section 2). That part of the knowledge which could be formalized (e.g. the knowledge about structural design), was, together with the methodological drafting model, formalized and represented using the frames-representation formalism. Most knowledge was represented in simple frameslot procedures regarding hierarchical and referential relations and serving to address relevant blocks of information, or support limited inferencing.
It will be evident that, by using hypertext as an implementation technique, it was not hard to transpose the methodological (norm-)frames-representation (within our design-step-model) into a hypertext network. We used the frames-representation specifically to model the hypertext network to our needs. For instance: in order to model the hierarchical links in the hypertext network, we used the methodological knowledge about drafting activities represented in the frames-network. In the same way we modelled the network's referential links and inference procedures. This enabled us to create a hypertext network which does not only provide very flexible information linking, but which also dynamically produces knowledge-based templates, and substantively as well as methodologically supports legislative drafting [Voermans et al., 1993b].
Continuing the discussion on the formal description of hypertext networks we will now focus on one particular type of hyperlinks, namely dynamical links. We will introduce a frames-formalism for their representation.
One of the prominent goals of the LEDA project mentioned in the introduction, was to make the information of the Recommendations accessible. Here we will discuss a form of dynamic access using the possibility of hypertext to create dynamic links from textparts to relevant recommendations. The aim, in line with the intentions of the Recommendations, is not to correct the text automatically but to indicate which recommendations have information on the subjects of the draft-text. The Recommendations give directions and guidelines which the draftsman can apply or ignore if this, by his own professional judgement, results in better legislation. The draftsman has to check his text himself on conformity to the relevant recommendations presented by the system. To achieve this goal there are two conditions to be fulfilled:
The first condition is one that can be answered fully. The Recommendations form a closed domain so the complete set of subjects of the Recommendations can be brought into a formal model. Our purpose is not to build a conceptual model of the contents of the Recommendations. The Recommendations for regulation contain a lot of vague norms that cannot be put in a formal model. Even if it could be done, such a model would only be useful for automatic reasoning and that is not our aim (see section 2.). The conceptual model must express the subjects of the Recommendations like a table of contents but then in a more formal way so it has handles for automated use. The concepts in which the recommendation-subjects are modelled, are representented with frames that are organised hierarchically in the shape of a tree. The hierarchical links are supported by the hypertext engine and can thus be followed. Generic concepts that have references to a large set of recommendations are high up in the hierarchy and concepts that refer to individual recommendations are placed at the bottom as the leaves of a tree.
The second condition gives rise to some problems. Ideally every subject in the draft-text on which the Recommendations have some information would be correctly recognised. With a complete conceptual model of the recommendation-subjects available this would still take full understanding of the text. Full understanding of natural language requires common sense and real-world knowledge and those are too complex and too extensive to ever be represented in a computersystem. We chose a more realistic approach that uses terms or word combinations as indicators for concepts.
PM_Level_Frame
[
/* Frame identification */
name : string;
number : integer;
/* Frame information */
status : {empty, editable, locked};
recommendations : set_of_string;
explanation : string;
activities : set_of_activity_frame
]
Activity_Frame
[
/* Activity identification */
act_name : string;
act_number : integer;
act_explanation : string;
/* Question - answer activity */
question : string;
answer : string;
/* Overview scheme activity */
scheme : set_of_characteristic;
/* Selection activity */
selection_question : string;
selection_answers : set_of_string;
selection_user_choice : string;
selection_recomm : set_of_string
]
characteristic = aspect x variant x string;
aspect = string x set_of_recommendations;
variant = integer;
Table 2: Generic frame: PM Knowledge-based templates
Concept_Frame
[
/* Network connections */
super_in_hierarchy : Concept_frame;
sub(1)_in_hierarchy : Concept_frame;
...
/* Tupelidentification of the concept */
terminology_concept : ConceptSet;
/* Match-information */
keyword_indicator : set_of_string;
concept_indicator : keyword_indicator;
/* Relevant recommendations */
relevant_recommendations : string;
/* Operation */
show_leaflet;
/* Leaflet */
leaflet : text
]
Table 3: Generic description of a Terminology frame
There are two dimensions to measure the quality of the concept-indicators: completeness and soundness. These measures are akin to notions from the discipline of information-retrieval: precision and recall. If the model with concept-indicators is complete then the recall of the conceptrecognition in the draft-text will be 100\% and likewise if the model is sound then the precision of the conceptrecognition will be 100\%.
There is a reversed proportional relationship between these two quality dimensions which manifests itself in the handling of vague concept-indicators (which indicate generic concepts) and ambiguous concept-indicators (which indicate, via multiple different concepts, the connecting generic concept). When we aim for completeness we will have to include these concepts too, accepting that the soundness of the model will decrease and vice versa.
With the conceptual model the system can, prompted by the legislative author if he has finished the drafting of a text, create dynamical links from concepts in the text to relevant recommendations in the database. The author can subsequently use these links to consult the recommendations that have information on the drafted text. To create the hyperlinks the parser not only searches the text for words or wordcombinations that can be matched with patterns in the database (stringmatching), but it also analyzes concepts in the text (using a form of the conceptual dependency method [Schank et al., 1981]) and matches them with concepts in the database (conceptual information retrieval). Some word(combinations) will indicate specific concepts which can be related directly to relevant recommendations. Others will be matched with the vague and ambiguous indicators of generic concepts.
In that case a link is created not directly to a recommendation but to a kb-template (leaflet) explaining the recognised concept. By the hypertext supported hierarchical ordering of the concepts we have incorporated the facility to let the user decide, by choosing a link to a more specific concept, what his exact intention was and thus access the relevant recommendations. This facility resolves soundness problems so we can safely aim for completeness and we may include vague and ambiguous concept-indicators.
Fig. 2: Resolving a vague concept indicator by user navigation through the concept tree
To handle the possible vagueness and ambiguity of words and to control the completeness and soundness for their use in the conceptual model we use a structured method for the modelling of recommendation-subjects and the assignment of concept-indicators. The method is generic in the sense that it is drawn up with knowledge of the possible forms of textnorms in general. It should therefore be possible to apply this method to any set of textnorms to build a conceptual model that can be used to facilitate the automatic access of those norms.
The method indicates how each individual recommendation should be handled in the conceptual model relative to it's character and how concept-indicators should be assigned. There are different ways to classify recommendations that can be more or less suitable for some purpose. A categorisation that is fit to differentiate on the way that recommendations should be treated in building this model is the relative distance to the text. The distance of a recommendation to the text is a metaphore for the chances and possibilities to express the relevance of the subject of a recommendation in a term or in wordcombinations.
Purely terminological recommendations --- that prescribe or forbid the use of one or more specific words --- are closest to the text, and contain the sought-after keywords themselves. For recommendations that regulate terminology, without mentioning a specific word, it is less straightforward to find suitable keywords. These recommendations seldom contain concrete indications about usable keywords. We will call them discretionary recommendations. A third group of recommendations prescribe no terminology whatsoever, but do nonetheless contain information, that is relevant to the author of a sentence or clause, i.e. recommendations that regulate the meaning rather than the form of a sentence or clause. We will call these recommendations substantial.
In order to disclose discretionary and substantial recommendations we have introduced a concept between the text and the Recommendations. Every recommendation that holds important information for the author of a piece of text, is part of a concept and every concept can be discovered in a text by means of one or more keywords. In choosing these concepts we made use of the strong structure of the Dutch Recommendations; every chapter and section possesses strong central themes. These chapter- and section-themes are our main- and subconcepts. The coherence among concepts and recommendations can now in a straightforward manner be deduced from the hierarchical structure of the Recommendations.
The performance of this conceptual information retrieval method depends on the right choice of keywords to unveil concepts in a draft text. Purely terminological and discretionary recommendations are part of concepts, and for (some of) these we do already have keywords. These keywords we will use to disclose all the recommendations that belong to one concept. Substantial and discretionary recommendations will thus be disclosed by means of keywords that were derived from other recommendations, which is possible because together they form a thematical unity.
With the concept-indicators that we find thus, completeness is not guaranteed. A lot of terminological recommendations prescribe the use of one or more words. The 'forbidden' words however are not always mentioned and they are viable keywords too. This is not a big problem if the recommendations mainly are a formal codification of current terminology. A second limitation could be the following: there is no guarantee that every substantial recommendation has a terminological counterpart that can deliver the keywords. In the Dutch Recommendations these drawbacks do not or hardly ever occur, but if this is to be a general approach these issues have to be addressed more thoroughly.
The quality of the concept-indicators is good in terms of soundness. Because the keywords are derived directly from the recommendations, no problems are likely to occur here. Of course all recommendations belonging to one concept do have to be a thematical unity. If the recommendations have weaker structure than the Dutch Recommendations, this is a point of attention.
The dynamic linking system that we developed with this method is currently being tested by the Dutch Ministry of Justice with (provisionally) good results. During the development we ran into questions that need further research.
In order to generalize the approach, some research has to go into finding limiting conditions, under which it is safe to rely solely on keywords, found in the Recommendations themselves. As these conditions will not always be met for every recommendation, we need a secondary method for finding keywords. Thinking about this method concentrates for the moment on things like: examining recent regulations and juridical literature (using e.g. statistical methods), and consulting field-experts.
A second field of attention is the method of choosing the concepts or themes. The concepts determine which selection of recommendations is offered to the author as a coherent unity. In order to make a selection that is useful to the author, questions relating to the choice of concepts have to be answered; these questions include: In what way does a professional work?, What are the problems he encounters?, In what coherence do they arise? and What help and information can the recommendations offer him?
A well recognised theory in the field of computer science is that the nature and the characteristics of a job determine the type of computer support that it is fit for. An intuitively appealing rule which is present in most systems development methodologies and which is equally valid for the field of legal informatics.
Legislative reasoning by nature is largely information based, and requires use of world knowledge and common sense, abilities which are exclusive to human beings. Complete representation of legislative knowledge is, if not undesireable, only possible for small and strictly bounded domains. Computer support should therefore be aimed at disclosure of relevant information and offering tools for problem exploration. By task oriented modelling of legislative reasoning we can narrow down the wide body of available information to strictly that information that is relevant in the given context and we can offer the user task specific tools that are powerful enough for representation of his ideas.
Hypertext offers the combined facilities of building a tailored information environment for specific kinds of legislative reasoning and a powerful autoring tool for problem exploration.
The hypertext based Legislative Design and Advisory system LEDA is currently being used by legislative experts at the Dutch Ministry of Justice in test settings and we have already registered enthousiastical reactions. Our research extends in the field of formal design of hypertext networks, the use of linguistic instruments for concept recognition and legal knowledge representation.