This paper describes Blawx, a prototype web-based user-friendly Rules as Code tool, powered by a goal-directed answer set programming. The paper briefly describes Rules as Code, and introduces desireable qualities for Rules as Code tools. It provides justifications for Blawx's implementation of the Google Blockly library, and the s(CASP) reasoning system. It then provides a step-by-step tour of how Blawx allows a user to generate an answer set program representing their understanding of a statute, and use that encoding to power an application. The paper concludes with a brief discussion of the current short term and anticipated long-term development objectives for Blawx.
https://www.lexpedite.ca/ (J. Morris)
© 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). CEUR Workshop Proceedings • sophisticated reasoning features, to ensure the representations are as useful as possible • natural language explanations for conclusions linked to legislative source material, to earn trust of experts, developers, and users • ease of integration with other technologies, to increase the value of encodings and promote adoption
While there are a number of commercial and open source oferings with some of these qualities, several of which have inspired aspects of this work3, Blawx is to my knowledge the only ofering with all of them.
Since late 2021, development of Blawx has been guided primarily by the needs of the Rules as Code Programme at the Canada School of Public Service, which is supporting departments across the Government of Canada interested in exploring the Rules as Code methodology.
Blawx features a drag-and-drop interface in which an answer set program is represented by ”blocks” which connect to one another in a way that is visually similar to connecting pieces of a jigsaw puzzle.
Blawx’s visual interface is developed using Google’s Blockly library4, which is based in research on teaching programming to children, and utilized in very popular pedagogical tools such as Scratch5. The puzzle-piece interface reduces the types of syntax errors that are possible compared to a traditional programming environment, and provides more immediate feedback with regard to others. This minimizes the cognitive load on beginning users who typically struggle to diferentiate between syntactic and semantic errors. 2.2. s(CASP) Work began in December of 2021 on re-implementing Blawx using s(CASP)6 as the back-end reasoner, moving away from Flora-27. The decision to re-implement in s(CASP) was taken after investigations into s(CASP) that took place in part at the Singapore Management University Centre of Computational Law and continued inside the Digital Experience and Client Data division of Service Canada’s Benefits Delivery Modernization programme. Those experiments demonstrated its feasibility as a tool for diagnosing and correcting legislative drafting errors 8, for powering expert systems9, and for dealing with the data types most commonly used in our target use cases10.
3See, in particular, AustLII’s Datalex at https://www.datalex.org 4Available at https://github.com/Google/Blockly 5Available at https://scratch.mit.edu 6Arias, Joaquin, et al. ”Constraint answer set programming without grounding.” Theory and Practice of Logic Programming 18.3-4 (2018): 337-354.
7Available at http://flora.sourceforge.net 8Morris, Jason. ”Constraint answer set programming as a tool to improve legislative drafting: a rules as code experiment.” Proceedings of the Eighteenth International Conference on Artificial Intelligence and Law. 2021. 9Docassemble-L4 is available at https://github.com/smucclaw/docassemble-l4 10Presented at Programming Languages and the Law 2022 workshop as ”Prevalence of Expression Types in Legislative Text”, see https://popl22.sigplan.org/details/prolala-2022-papers/11/Prevalence-of-Expression-Types-inLegislative-Text
These experiments revealed several potentially important benefits from using s(CASP) that were not anticipated. In addition to the critically important feature of natural language explanations for conclusions, experience showed that multiple stable models, dual programs, and abducibility provide the legal knowledge engineer with a vastly better understanding of what their encoding actually means in the development phase, and in turn much greater confidence that the semantic meaning of the code is consistent with their interpretation of the rules.
The user experience of using Blawx for Rules as Code involves providing the legislative text, encoding the legislative text in a structurally-isomorphic way, testing those encodings, and then integrating those encodings into other systems.
In order for encodings to be structurally isomorphic to the law from which they are derived, Blawx needs to be aware of the structure of the source law. Akoma Ntoso11 (also known as LegalDocML), is used to represent legislative texts inside the Blawx system. Open source, user-friendly editors for LegalDocML are rare, and those that do exist use styles of legislative drafting unfamiliar in the Canadian context.
For that reason, I created CLEAN12 (Canadian Legislative Enactments in Akoma Ntoso), which is a markdown-like plain text syntax for legislative text, with tools for converting it into Akoma Ntoso. Blawx allows the user to specify legislation in CLEAN. The interface is shown in Figure 1.
Once the law has been entered, the user is presented with a side-by-side coding environment with the text of the law on the left, and the Blawx coding environment on the right. The text of the law is annotated with radio button selectors, which are used to select the hierarchical section of the law for which the user wants to generate code. If the user wants to encode parts smaller than a numbered section of the law, CLEAN allows for a span syntax to divide a block of text into arbitrarily deeply nested named parts, which Blawx will treat as though they were numbered sections of the law. This interface is seen in Figure 2.
The user selects one section of the law, and represents that section of law in the coding window. Ontology (categories, attributes, and objects) declared in any section of the law become available for use in all sections of the law, eliminating the need to redefine terms that have been defined elsewhere.
When sections of the law are expressed in a non-monotonic way, such as defaults and exceptions, the user has access to language elements that allow them to attribute a conclusion to a section of law, specify which conclusions conflict, and specify how to resolve those conflicts as 11Palmirani, Monica, and Fabio Vitali. ”Akoma-Ntoso for legal documents.” Legislative XML for the semantic Web. Springer, Dordrecht, 2011. 75-100.
12Available at https://github.com/lexpedite/CLEAN between two rules. In this way, the user is able to maintain structural isomorphism regardless of whether in the legislative text the conflict is explicitly resolved in the default (e.g. ”subject to following sections”), or in the exception (e.g. ”despite section 3 above”). These statements are resolved using a very simple argumentation theory inspired by the approach taken in Flora-213. An example of an encoding using defaults and exceptions is shown in figure 3.
Once the law has been encoded, the user is able to create ”tests”, which are another coding environment with access to all of the ontology declared in the encoding of the law itself. Here, the user can create a fact scenario and a query, have the reasoner run that query, and display the results. If the conclusion is one to which the defeasibile argumentation theory applies, there is a version of the query that allows the user to ask whether a given legal conclusion holds, despite being defeasible. The user can also ask whether a conclusion held according to any section of the law. An example test is shown in Figure 4.
13Wan, Hui, et al. ”Logic programming with defaults and argumentation theories.” International Conference on Logic Programming. Springer, Berlin, Heidelberg, 2009.
s(CASP) code is generated from the Blawx code, and saved to the server. When the test is run, the server collects all of the s(CASP) code for the sections of the law, adds the s(CASP) code generated for the test, and adds libraries that are used for defeasibility and date calculations. The query is then sent to SWI-Prolog14 using the s(CASP) library15, and the results are displayed to the user. Models with the same bindings are grouped into an ”answer”. Each ”answer” shows its bindings (if applicable) and one or more ”explanations”, which are a collapsible tree view of the natural language justification provided by the s(CASP) reasoner. When the defeasibility elements of the encoding refer to sections of the source law, those sections are highlighted in the text of the explanation. Hovering over the highlighted text display the text of the legislative section to which they refer. An example of the answer display is shown in Figure 5.
Each test that is created in Blawx automatically generates three web API endpoints designed to allow the user to collect information about the ontology defined in a given test, and to run the included queries with or without information about what additional inputs might be relevant to 14Avaialble at https://swi-prolog.org 15Available at https://github.com/JanWielemaker/sCASP ifnding stable models for the query. These interfaces are designed to allow the Blawx encoding to be able to power expert systems and other tools elsewhere on the web. As a demonstration of the use of these interfaces, an example expert system called ”BlawxBot” is available. Running BlawxBot displays a chatbot-style user interface where the user is prompted to provide the details of a relevant fact scenario. As soon as a stable model exists the interview is terminated and the result is displayed to the user. An example of the BlawxBot interface is shown in Figure 6.
Work on Blawx in the near term will continue to be guided by the needs of the Canada School of Public Service and its client departments. Currently, our priorities are: • Complete BlawxBot expert system chatbot interface • Enhance documentation, examples, and training materials
We then hope to test Blawx extensively with real-use world cases, and use feedback from the users to guide future development.
The Rules as Code movement suggests that legal knowledge representation has great potential benefits for policy design, legislative drafting, public administration, compliance, and other endeavours. Our experiments have shown that s(CASP) has great potential as a tool for statutory legal knowledge representation. Blawx is an attempt to facilitate the use of s(CASP) by nonprogrammer subject matter experts for Rules as Code tasks, to explore how we might unlock that potential. Recent months have seen a focus on developing a prototype which is now almost complete. Work will continue inside the Government of Canada testing that prototype against real-world use cases to improve both the tool and our understanding of the Rules as Code methodology generally.
My thanks to Pia Andrews, and to the DECD team at the Benefits Delivery Modernization Programme in Service Canada. Thanks also to the visionary leaders at the Canada School of Public Service who are leading the world in commitment to exploring the Rules as Code approach in the public sector. In particular, thanks to Marty Perron of CSPS, who has provided valuable direct feedback on Blawx’s development.
Thanks also to the leadership and staf of the Singapore Management University Centre for Computational Law.
Thank you also the developers of s(CASP) for their support and assistance, and to the organizers of this conference for their interest in my work in the legal domain.
Research having occurred at the Singapore Management University Centre for Computational Law was supported by the National Research Foundation (NRF), Singapore, under its Industry Alignment Fund – Pre-Positioning Programme, as the Research Programme in Computational Law. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not reflect the views of National Research Foundation, Singapore.