Overview
a. Dandi is envisaged to be an installable desktop application to be used by participants who have to deal with law - namely lawyers, Chartered Accountants, judges, law enforcement agencies, government legal cells, law faculty, law students etc. Dandi shall provide its users with a tree (or graph) like structure of the legal act the user might be interested in. Dandi shall also parse the judgments of various courts and appellate authorities and link them to the appropriate section of the legal act, the judgment might be related to.
b. Dandi is meant to assist the practitioner of law to understand the structure of statute (in a visual graph like manner), the chain of dependencies of various sections upon one another, and the relevant judgments, circulars, notices, and instructions related to the statute, in order of their importance.
What Exists
a. Currently, in the domestic market, there are specialized products available for various legal acts (hereinafter referred to as ‘act’ or ‘acts’ as the context may be). These specialized products, usually pertain to one single act (say Income tax Act); provide functionality of view of the various sections (which is not really of much value since the bare act is anyway available with all practitioners in paper or electronic form or over the internet) and search of case laws (judgments of various courts) based on keywords as well as based on relation to any desired section of the act.
b. Prior to the popularity of softwares the case laws (those which were considered important) were (and also are) published as series of compilation, running into 10-20 books per year, by various houses. Some of the most popular publications for Income tax purposes are ITR, CTR, Taxmann, TTJ etc. The practitioners used to subscribe to the same on yearly basis and thus used to have access to the various court pronouncements in the country. The case laws are uniquely referred to, in all legal texts, based on the volume and page number of the publication, e.g. 23 ITR 439, 129 CTR 28 etc.
c. Softwares such as Taxmann, Taxpert, and some more, have penetrated the Indian market and instead of shipping physical books, they make the case laws, along with search facility, available at the user’s machine, through installation of software from the CD. In other words, these products, have the case laws packed in a CD which is then purchased and installed by the user, on an yearly basis, for a charge ranging between Rs. 5000-10,000 per annum. These softwares also update themselves periodically, when connected to the internet.
d. The USP of these products lie in their database of case laws and more than that in their search facility. Apart from this they create a synopsis of the case law (i.e. a one paragraph summary of the entire judgment as per their own understanding). But more importantly in these products the linkage between the case law and the relevant section and keyword is static and performed manually. In other words, a subject expert reads the case law and links the judgment to a given section (or sections) of a given statute, and perhaps also links it to the set of keywords the case law is related to.
e. It is presumed that similar software may also be available for other acts and departments.
Dandi’s USP
a. It may be mentioned that Dandi is envisaged to be a generic engine which takes as input any act, parses it, and creates a tree (graph) out of the same. Dandi also takes as input a case law, parses it, and links it to one, or more than one, nodes of the tree created earlier. Thus Dandi engine is supposed to be act neutral and generic in nature.
b. If the engine is successfully created, it may very well create an object oriented structure for any act of the world, let alone India; and in this sense would be highly scalable.
c. It may also be mentioned, that none of the existing products provide granularity, smaller than a section. That is, the least count of the exiting products is section and it does not identify the sub section, clauses, or proviso. In other words, the break up of act by the existing products is liner – the act divided into a list of sections. Dandi intends to create a tree instead of flat structure.
d. Currently, the existing products rely on manual expertise for the association of the case law to a given section. If rudimentary natural language parsing can be done for English then this feature could be automated in Dandi. In other words, the case laws need to be fed to the Dandi engine, and it shall be able to automatically link it to the relevant section of the relevant act. Again a feature, which would make the product highly scalable and marketable across countries.
e. If the Dandi engine is capable of parsing the case law text and identify the references to other case laws in the text, then a ranking algorithm like Google Page Rank can be implemented in the world of case laws to determine which case laws are more important.
f. It may be pertinent to mention here that in the world of case laws, one judicial pronouncement, more often than not, relies (and/or counters) upon a series of case laws before reaching the verdict. However, there are certain case laws which are of greater importance (either because of being from a higher court or because of their discussion) and are cited again and again by the judges in support of their order. Thus there is an intricate reference and linking between case laws, much like the pages on world-wide-web. Currently subject experts take the call as to which case law is important and which is not. The subject experts, themselves, inter-alia, rely on their own knowledge of how frequently the given case has been used in other judgments.
g. The idea of Dandi is to formalize this algorithm in the engine and apply Google Page Rank (or a variant) to determine the relative importance of the case laws. The assignment of weight to a given case law, if at all, in the existing products is only known to be manual and static so far. The same can be automated without much difficulty, provided Dandi is able to perform the requisite degree of natural language parsing.
h. If the requisite degree of language parsing can be performed, which the current author believes can be done, then a Google equivalent can be created in the world of law.
Main Features
a. Tree Structure - All acts, through out the world, are divided into chapters, sections (or sometimes called articles), sub-section, clauses, provisions etc. Dandi, shall take as input, in plain text format, any act and create out of the same an object oriented structure by identifying all chapters, section, sub-section, clause, sub-clause, provisions etc. The above parsing might require one to supply a short meta-data for any given act specifying the nomenclature used by the same (whether section or article etc). This task, however, does not require natural language parsing. The parsing required for the above purpose would be akin to the parsing employed by compilers to check syntax; though it needs to be very robust to accommodate all types of acts.
b. Once a tree is created out of an act, the Dandi engine may read the text of the specific section to find out the other sections which might be referred to in it. This requires legal language parsing. In the opinion of the current author, legal language, in itself, is close to symbolic logic and, for the limited purposes of extracting intra and inter act references, may not present a bottleneck.
c. The two level of parsing, if successfully done, shall present a logical graph of the act, which a lawyer develops only after years of practice. The said graph shall depict dependency of one section on another and thus may help in identifying and appreciating all pre-requisites clearly, and in a visual manner.
d. The said tree structure may also allow referencing between acts, as one act referring to another is common, and may provide seamless navigation between acts.
e. In a way, in the world of acts and sections, Dandi intends to create a ‘Wikipedia’, where whenever a reference is made to another object (a section of any act), a link is automatically created for seamless navigation. Currently, such facility, wherever provided in the current softwares, are hardlinks created manually.
f. The USP of such tree would be to bring about a genuine enhancement of understanding of the reader on the subject matter. It might also prove immensely useful to the teachers and students of law around the world, as well as the legal draftsmen.
g. Apart from the same, such tree may also be able to identify sections, or sub-sections which are not referred to from anywhere in the act and may have become superfluous. Though that might be something of interest only to the law-makers.
h. Amendments – Every year the acts get amended i.e. certain sections get repealed, some get added, and some get their wordings changed. The Dandi engine shall be able to take as inputs the amendments, and appropriately modify its graph. Thus amendments would be akin to ‘patches’ of updates, applicable from a given period.
i. Implementation of the above would also allow the user to have a view of the act at any past point. It may be mentioned that invariably disputes in court of law, pertains to some past year, and the argument and judgment has to made in the light of law prevalent at that time. With object oriented view of the act the old act, and the corresponding contemporary acts, would become retrievable by merely specifying the date of interest, thus adding great value to the users.
j. Case Laws – As mentioned earlier, the case laws are pronouncements by various courts, tribunals, and other appellate authorities. The important one among these are listed in the popular compilations such as CTR, ITR, etc. The case laws itself are judgments in plain English and does not constitute legal language in the same manner in which statutes are. However, it may be possible to parse the text and identify references to other case laws and provisions of act.
k. The above task would require natural language parsing (though not to the extent of deciphering semantics) with the limited intent to identify references to statutes and other case laws. The nature of parsing (or more enhanced form of pattern matching, if one may please) itself might not present great difficulty and should be similar in complexity to the task of identifying references within the statute, as envisaged while creation of the intra-referenced tree structure for an act.
l. Assuming, that such restricted form of parsing can be done, the Dandi engine may be able to create a web of case laws and acts; where each case law references some case laws and some provisions of the act. Also, each case law, would in turn be referenced by some other case laws.
m. Once such web is created, the Dandi engine, through an iterative process, applying Google Page Rank algorithm, or a suitable variant, revise the weight attach to each case law and come up with a ranking. In may be pertinent to mention that even a simple strategy as to how many judgments a case law has been used in, shall yield good results for deciding the relating importance of case laws.
Creation of Database
a. Creation of database of acts and case laws itself is an involved and costly affair. However, if the Dandi engine is robust and efficient (in terms of accuracy and not in time) then the database can be built up without incurring high cost.
b. The feeding of act and its meta data is a one time process and may not present cost issues. The case laws themselves are numerous and available in public domain. However, if the engine is smart and only feeding of bare text of case law to the Dandi engine is required, without application of mind, then the database can be quickly build up at low cost using semi-skilled workers.
c. Apart from the above, the creation of the database may also be accelerated by letting users feed new case laws, somewhat like Wikipedia, subject to certain checks and incentives.
d. At any rate, if the database itself creates problem, an initial tie-up with one of the less performing companies can be done in the initial phase.
Pitfalls
a. The only bottleneck as envisaged by the current author constitutes of the extent to which parsing can be performed. There are three level of parsing envisaged in the product – a) syntax parsing of the bare act to create tree structure, b) Legal language parsing of the provisions of the act to identify references to other provisions of the same or other acts, c) Natural language parsing of the judgments to identify references to statutes and other case laws.
b. A robust implementation of the above three types of parsing is mandatory to have a truly scalable and global product.
c. With a good engine, creation of database shall be quick and cheap. In fact, with a good engine, the same can be very quickly developed and marketed in other countries.
Way To Go
a. It may be pertinent to mention that for all countries the Income tax Act is order of magnitude times more complex than any other statute. If the engine tests and works for Income tax Act, it is likely to work in all statute environments.
b. The development for syntactical parsing of the act (and Income tax Act may be taken for the purposes of development and testing) and the development for language parsing, both the statute and case law, can be taken simultaneously, with may be some agreed class structure. The parsing of natural language would be in the nature of a research project in Artificial Intelligence, and assistance from a professor might as well be sought.
c. Once the two parsing engines of Dandi are ready which creates objects and links out of plain text inputs, the rest can be easily implemented.
Log in or sign up for Devpost to join the conversation.