Legal Information Retrieval by Computer:
Applications and Implications
Colin Tapper *
It is about fifteen years since Professor Horty and his team
at the University of Pittsburgh decided that computers should be
enlisted to keep track of the great and constant expansion of
legal information. During these years the computer has been
applied to modem statutes and to ancient cases; by universities,
by professional bodies, by governments and by entrepreneurs;
and everywhere from Austria to Australia. Every combination of
keywords, abstracts, indices and original text has been tried. They
have been stored in a variety of formats for searching in any
number of ways to yield output in a myriad of different forms.
Enormous sums of money have been spent and aeons of man-years
consumed in the creation of notations, languages and programmes.
Even in this relatively short period, it can already be said that
no one can hope to be fully informed about the developments
which have taken place. Certainly no one article can purport to
delineate their major features. But the time is appropriate to
draw back, to consider what has been achieved, to evaluate it, and
to ask what should be done next.
I APPLICATIONS
The last fifteen years have witnessed a flourishing of computer
techniques in their application to the law. A number of interesting
experiments have been conducted and a few practical systems
have been brought into operation. It is convenient to deal with
each one of these three elements separately, bearing in mind that
they do not exist in self-contained compartments since all opera-
tional systems necessarily apply some techniques and most have
been tested experimentally.
(a) Techniques
Fifteen years ago computers were very limited in capacity and
very slow in operation. Input was usually by punched card, no
auxiliary storage was available, and output was usually by line
* Fellow and Tutor in Law, Magdalen College, Oxford.
1974J
LEGAL INFORMATION RETRIEVAL BY COMPUTER
27
printer. The computer would accept linguistic information only
in upper case, and could store it only serially. It was therefore
remarkably far-sighted of Professor Horty to think of using these
techniques for handling the very large volume of data contained
in the Pennsylvania statutes, particularly since he proposed
to
input the full text of those statutes, and not merely an index to
them. In a sense this decision was forced upon him since his
basic goal was to identify insufficiently indexed sections of the
statutes which dealt with health law with a view to compiling
a manual on that subject. The technique which he pioneered
involved the preparation of the full text of all the statutes in their
original language, and the creation within the computer of a complete
concordance of all, except a few common words. The retrieval
system then relied upon the detection of the occurrence and
conjunction of particular words in the text to identify those parts
which related to the required subject. Because of the technical
limitations of the computers available to Horty, it was most
economic to aggregate a number of searches together and to perform
them all at one time, a technique known as batch processing.
It was a remarkable achievement to make so many innovations
in the techniques for finding legal information simultaneously,
and it is hardly surprising that it took some time for them to
become fully understood and accepted. It is only now that significant
variations of method are beginning to be proposed. One reason
for this delay is paradoxically that the pace of technical improvement
has been increasing. As a result, a great deal of the available
expertise has been devoted simply to adapting Horty’s techniques
in the light of these technical innovations. In a very short period
of time, one has had to learn how to deal with optical character
recognition for input, random access auxiliary stores and terminal
systems permitting direct interrogation of the system by the user.
Optical character recognition is a technique which speeds up input
by allowing the computer to accept material directly from the typed,
if not printed, page. It should be noted here that the possibility
of accepting information directly from the printed page without
the intervention of a human transcriber will create problems of
a new order of magnitude, and will be considered later. The
availability of random access auxiliary stores was a thoroughly
liberating advance since it meant that many operations which
had previously cluttered the central store of the computer and
thus limited its performance could be performed elsewhere in
the system. This resulted in an enormous increase in both capacity
and efficiency, although it is interesting to note that some now
feel that the transition from serial to random access has not been
McGILL LAW JOURNAL
[Vol. 20
wholly advantageous.’ The most obviously revolutionary of the
new factors mentioned above is the facility for the use of terminals
which, together with the availability of multi-user capabilities, has
necessitated a complete reappraisal of searching techniques. In the
more sophisticated systems, terminals now comprise keyboard,
video screen for the display of information and an associated device
for preparing hard copy. The great change (not entirely in
accordance with Horty’s original philosophy) which this brings
about is to put the user into close contact with the progress of
his search while it is still going on. This enables him to control and
modify it in the light of his interim results.
While
it is true that in
the past fifteen years
linguistic
applications of computers have increased in all fields and that
in consequence standard programmes for such applications have
become
increasingly available, one of the results of Horty’s
initiative has been that legal systems have been in the forefront
of these developments and have had to bear much of the strain. This
has led to a certain distortion in emphasis. More time has been
spent on getting the computer programming right in legal informa-
tion retrieval systems than on getting the legal techniques right,
and this may well have been unfortunate.
Professor Horty’s original application was to the Pennsylvania
statutes. A further area in which effort has been made is in the
adaptation of the techniques to the demands of other types of
legal material and to its expression in other languages. A number
of different workers soon began to apply these techniques to patent
applications, to administrative rulings and above all to cases.
The problem was to determine how far and how effectively these
methods, depending as they do upon form, could be applied to
legal materials expressed in forms different from that of American
state statutes. Thus when an attempt was made in the early 1960’s
to test Horty’s techniques on British public health statutes it was
found that none of his test questions were satisfactory because
of the difference in the level of generality in the two systems.
This is even more true of non-common law systems where the
functional balance between statutes, cases and other materials is
quite different, and is naturally reflected in differences in the
form of those sources. One general similarity, however, is that
legal information is at least everywhere very largely linguistic,
and it has thus generally been unnecessary to grapple with the
I See Thomas, “GIPSY – General Information Processing System for Okla-
homa”, in American Bar Association Automated Legal Research (1973), 119.
1974]
LEGAL INFORMATION RETRIEVAL BY COMPUTER
29
problems of computer graphics, although land registration systems
and systems for dealing with treaties which often incorporate maps
are on the verge of such problems. But even the restriction to
language provides only limited relief in relation to languages like
Hebrew which differ so radically in their structure from that
of English?
(b) Experiments
Although Professor Horty was originally conducting an essentially
practical exercise, namely the production of a manual, his thesis
of the possibility of computer application had, to be tested
experimentally. It was necessary to discover whether or not word
occurrence was an adequate surrogate for meaning, and if possible
to measure the divergence between results achieved by reference
to both occurrence and meaning. In this regard Professor Horty
was at a disadvantage in two ways, one practical and one theoretical.
The practical disadvantage stemmed from the statutory nature of
his material. It is notorious that statutes are extremely difficult
to index, and thus the conventional yardsticks for the retrieval of
the statutory sections relevant to a given problem were more than
ordinarily defective. It was indeed this very phenomenon which
had given the original impetus to the idea of using computers.
It followed that it was extremely difficult to measure the results
of the experiment accurately, and that such results could not
necessarily be extrapolated to other better indexed areas. It
is
also the case that searches for solely statutory material are rare
in legal practice, and that any testing of a system simply by
reference to such searches is unlikely to be typical. The theoretical
disadvantage was that Professor Horty was then unaware of the
techniques devised by Professor Cleverdon at Cranfield in Great
Britain for the measurement of the performance of information
retrieval systems. Cleverdon’s method involved the definition of
and distinction between the two concepts of recall and precision.
Recall connotes the proportion of relevant documents retrieved
to the total number of relevant documents in the collection searched;
precision connotes the proportion of relevant documents retrieved
to the total number of documents retrieved. It was Cleverdon’s
thesis that recall and precision tend to be in inverse relation to
each other. Thus it is possible to maximize recall by retrieving every
2Fraenkel, “Full Text Document Retrieval”, in Proceedings of Association
for Computing Machinery Symposium on Information Storage and Retrieval
(1971).
McGILL LAW JOURNAL
[Vol. 20
document in the collection since if all the documents are retrieved
it follows that all the members of every sub-set, including all
the members of the sub-set of relevant documents, must also be
retrieved. But this would simultaneously maximize imprecision.
Conversely imprecision can be minimized by retrieving no document
at all, since again this would entail that no irrelevant document
was retrieved. This strategy, however, would simultaneously
minimize recall since no relevant documents would be retrieved
either.
A further experimental problem is
that the use of these
measurement techniques assumes an objective assessment of rele-
vance. Professor Horty attempted
to solve this by graduating
retrieved documents into one of three categories of relevance, but
he does not indicate how far the objectivity of the ascription of
documents to categories was tested. This particular problem came
into prominence in one of the early attempts to apply Horty’s
techniques to case law in an experiment conducted by a joint
team from the American Bar Association and International Business
Machines Ltd. It was found that divergent judgments between
different testers was so acute that it became quite impossible to
establish sufficient agreement to give any statistical validity to
the results
My own experimental work attempted to utilize the Cranfield
concepts for case law while avoiding the pitfalls of artificial search
formulation and the subjective assessment of relevance. 4 Case law
is in fact uniquely advantageous in these respects. It is possible
to make good use of its characteristic feature of internal citation.
Judgments normally cite other judgments which are relevant and
patterns of citation can be traced easily in special indices. It is
thus possible in relation to any segment of past case law to
establish a list of all later cases which cite a case comprised in that
segment. The facts of such cases can then be used as problems for
the purpose of the experiment. This procedure has two advantages.
First, since the test represents an actual decided case it is clear
that the problem is real and not just an academic exercise; if
enough cases are aggregated, the chances of the results being
extrapolated into case law generally are high. The second advantage
sEidridge, “An appraisal of a Case Law Retrieval Project”, in Proceedings
of Computers and Law Conference at Queen’s University (1968).
4 Tapper, Feasibility Study of the Retrieval of Legal Information from Two
Types of Natural Language Text, Office for Scientific and Technical Informa-
tion Research, Paper 5062 (1969).
1974]
LEGAL INFORMATION RETRIEVAL BY COMPUTER
31
is that the fact of citation itself represents an objective decision
about the relevance of the cited case to the problem: it is made
independently of the experiment by a professional judge in the
course of his ordinary work. On the basis of these advantages it is
possible to establish a firm low point on the precision scale (the
proportion of such cited cases to the total number of cases
retrieved) and a corresponding high point on the recall scale (the
proportion of such retrieved cited cases to the total number of
such cited cases). But an experiment based solely on such criteria
would be somewhat arid since it would assume that the current
practice of citation was not only accurate in its judgment of
relevance (which is plausible), but also completely exhaustive (which
is not). It is thus necessary to court the dangers of a subjective
assessment of relevance. In my own work this was avoided by
adopting a threefold distinction into cases relevant without doubt,
cases irrelevant without doubt, and all the rest. This procedure
was designed for speed, and by eliminating all doubtful cases,
for accuracy where a positive judgment was expressed.5
(c) Operational systems
Professor Horty’s system was always intended to be a working
one, and after a decade of development it became the basis of
a series of services offered by Aspen Systems Corporation, now a
subsidiary of the American Can Company. Although these services
have diversifed over the years, their heart has always lain in the
exploitation of the original data base, statute law. It became
apparent in the course of collecting state statutes in order to build
up the data base that there were significant difficulties and delays
within the various states in the compilation and up-dating of their
statute books. It was also discovered that many state statutes
contained anomalies and long-disregarded provisions. This inspired
Aspen to offer more comprehensive services involving the collection
and collation of state statutes. This made good commercial sense
in the United States where state governments were substantial
and creditworthy customers, and ones who might well provide
further markets for statute-based services. Of course in a free
enterprise economy Aspen was not long alone in the field, and
rival services began to be provided, notably by the Data Retrieval
Corporation of America, and by the state governments themselves,
the latter often at ihe prompting of salesmen from International
5 For a fuller account, see Tapper, Computers and the Law (1973), ch. 6.
McGILL LAW JOURNAL
[Vol. 20
Business Machines Ltd., which helped a number of the pioneer
states.
Under the stimulation of this competition the orientation of the
services changed from computer based legal information retrieval
to computer based preparation and printing of state legislation,
though information retrieval has never been completely forgotten.,
The advantage of sticking to statutes was that the data base
was relatively small compared to the bulk of case law and other
secondary sources such as encyclopaedias, books, articles and
opinions. Thus far the problem of the accurate transcription of volu-
minous archival material has constituted the principal stumbling
block to the transference of Horty’s full text techniques to case law in
a commercial environment. An obvious alternative is to use some-
thing less than the full text for case law. It can indeed be argued
that statute law is unique in the austerity and authority of its
language. Case law is relatively more discursive and diffuse and
much more difficult to abridge either into index terms or an
abstract. Index terms and abstracts do indeed regularly appear
as part of the published version of a case in most series of reports,
and index terms do operate as search tools within the conventional
system. In the United States and in some parts of other legal
systems 7 such shortened versions are linked into formal hierarchical
structures like the West Publishing Company’s key number system.
In other codified systems of law it is possible to employ the
organization of the code to supply a structure for other primary
materials. The obvious tactic in these circumstances is to use such
a structure or scheme of indexing as the data base for the
computerization of case law. Such an approach has indeed been
adopted by such various organizations as Law Research Services
Incorporated of New York, and CREDOC in Belgium. Where an
index is interposed between the user and the material there is
less need for continuous control over the search by the user, but
more need to ensure that the search terms match the search needs.
These considerations lead naturally to the use of batch processing
systems with the formulation of the search reviewed, if not under-
taken, by the organizations responsible for devising the system.
The relative decline in the cost of handling and storing large
quantities of information has, however, led others to keep to Horty’s
6For a full survey, see Elkins, Survey of the Use of Electronic Data
Processing by State Legislatures 2d ed. (1971).
7 The maxims collected by the Supreme Court of Cassation in Italy, for
example.
1974]
LEGAL INFORMATION RETRIEVAL BY COMPUTER
33
original philosophy, and to prepare systems for handling case law
in full text. Such systems are probably furthest advanced in North
America.8 The largest and most widely available to ordinary lawyers
is the LEXIS (formerly OBAR) 9 system in the states of Ohio and
New York. This system incorporates a full text case law data base,
permits freely structured searching in ordinary language, displays
results upon a video screen situated in the lawyer’s office and
allows any amount of modification to the search while it is being
conducted. Technically it is a very smooth running and sophisticated
system. Lawyers, however, have been slow to take to it, even
in Ohio where it was devised and where it has been on offer
longest. A possible explanation is that the system does not yet
do enough of the elementary thinking for the lawyer. For example,
there is as yet no automatic thesaurus to suggest synonyms,
antonyms, particularizations, generalizations, or grammatical or
orthographical variations of the chosen search terms. Further,
the search logic is restricted to Boolean operators supplemented
by qualifiers for distance and direction of conjunction. Other
systems have been designed to overcome some or all of these
defects; but so far none has been offered long enough or widely
enough to practising lawyers for it to be possible to determine
how much more popularity they will inspire.
It seems then that in the last fifteen years much work has been
done but relatively little has so far filtered through to the practising
lawyer. The original techniques devised by Professor Horty have
been improved and adapted, but not abandoned. Statutes are in
some sense being handled by computers in a practical environment,
but case law is not. In the second part of this article some
implications to be drawn from this state of affairs will be canvassed
and possible new directions for further research suggested.
II IMPLICATIONS
It
is arguable that the lack of response by the practising
profession to computers for the retrieval of legal information,
which is indicated as much by its failure to develop better systems
8 While the STATUS system developed by Dr Niblett at the Cuiham Labo-
ratory in Great Britain is used for statutes, it is quite capable of handling
case law and exhibits many of the features of operational North American
systems.
9 This system is described on the basis of the latest published material but
it is possible that it has by now developed beyond the stage described in the
text.
McGILL LAW JOURNAL
[Vol. 20
as by its failure to use those which are already provided, shows
that there is no need for such facilities to be provided. After all,
despite the direst forecasts and gloomiest projections, the legal
profession still does cope with the increase in material and there
is no obvious decline in its standards. It may be the case that
relatively few practising lawyers do enough research to justify
investment in a computer orientated service.'”
It might indeed
make more sense to concentrate on the use of the computer to
assist more directly in the solution of legal problems, but that
is outside the scope of this article. But even if consideration is
limited to legal information retrieval, there is a further stumbling
block which may have contributed to the difficulty so far experienced
in getting commercial services off the ground. It is customary to
think in terms of searches for legal information. This is a vague
and misleading formulation. To start with, most systems provide
references to where information is to be found rather than informa-
tion itself, but this is a quibble. Of much greater importance
is the assumption that searches all share a common character
and that the sort of information which will satisfy one will also
satisfy another. This is fundamentally implausible. It has ‘already
been suggested that typical searches in statute law differ from
typical searches in case law, and that academic searchers are
often looking for something different from practitioners. Further,
each different sort of lawyer at each different stage in his work
in each different class of case in each different system of law
will need to search in a different way. Thus there is not one common
pattern of search but an almost infinite number. It is not obvious
that they all have enough in common to enable any one system
to satisfy them all. For example, the tax adviser who wants to know
whether a particular concession is still being allowed by the revenue
hardly requires the same sort of information as an academic lawyer
interested in discovering whether there are any analogies to the
requirement of the identity of the mental capacity required for
making a contract and consenting to the dissolution of a marriage
in other branches of the law. Again one sort of search may require
an answer in terms of the current law of a particular jurisdiction,
another may want only the law at some particular time in the past,
or at some other particular place. One search might require no more
than citations, another an indication of the volume of materials,
another an indication of their contents, and another suggestions
10 See Operation Compulex, Department of Justice, Ottawa (1972), 25, 26,
app. 2, chart C.
1974]
LEGAL INFORMATION RETRIEVAL BY COMPUTER
35
is perhaps conceivable that systems of
on how to proceed. It
such generality and flexibility could be devised to supply informa-
tion in many forms to a variety of searchers. But such a system
would be incredibly wasteful of resources unless need was spread
evenly and at random across the whole spectrum of possibilities,
which is intrinsically unlikely. It is clearly necessary first to
discriminate between the different sorts of searches which are
made, and then to discover how much need there is for each
of them. This is not, however, an easy task since current practice
will not necessarily give an accurate answer. It will itself be
influenced by the efficiency of currently available aids. Need may
be better measured in economic terms, but even this is not an
easy task when the comparison must take in lawyers in private
practice, lawyers employed by the government or commercial
organizations, members of the legislature, voluntary advisers, legal
academics, students and private citizens. But however difficult
the task may be, it is clear that it must not be shirked. We are
at present in the era of trial and error at random, but the sums
of money and resources involved are becoming so great that this
cannot continue long.
In what follows it will be assumed that some such appraisal
of need has been conducted, and that some legal information
retrieval system by computer has been found to be desirable.
In this event it will be necessary to improve the performance of
the systems currently being offered in various ways and to under-
stand the consequences which will follow from the availability of
such improved facilities. These two elements will be considered
separately.
(a)
Improvements
Technical improvements will continue to be made, programmes
will become more efficient and both will occur in ways which
are partly the continuation of current trends, and partly un-
predictable innovations. Possible improvements in data capture,
searching and testing procedures will be discussed here.
(i) Data capture
As stated earlier, data capture has continued to constitute
the most serious financial and time consuming obstacle to the
development of large scale legal information retrieval systems.
There are really two separate problems: the preparation of archival
material and the assimilation of new material. So far as archival
McGILL LAW JOURNAL
[Vol. 20
material is concerned, the difficulty has resided in the inability
of optical character recognition devices to deal with the wide
variety of type faces to be found in the conventional law book.
It is likely that these difficulties will one day be overcome. However,
that will not bring all problems to an end; it will simply mean
that a different set will have to be faced. At present transcription
is carried out by men who apply their minds to the task and
who are able to follow simple instructions. If they are asked to
insert special characters at the end of each document they are
capable of doing so, and even without special instruction they
will be able to distinguish between hyphenation inserted by the
printer to help justify the right hand margin, and ordinarily hyphen-
ated words. The transcription device and the computer will be
unable to accomplish either. All additional symbols will have to be
specified and inserted at some stage in the transcription process.
Similarly very precise instructions will have to be devised to accom-
plish any necessary deletions or modifications of the text, such
as the relation of footnotes to the point in the text referred to.
Further, in documents which depend for their meaning on a
special format, such as schedules of repeals in statutes, or which
contain tabular material or diagrams, some method of treatment
must be prescribed. Finally, it will still be necessary to check the
accuracy of the text as transcribed since even if it is assumed
that the device will be completely accurate, it will still reproduce
any typesetting mistakes in the original.
Problems with new materials are likely to be less severe, and
are in some ways the converse of those described above. The
danger here is that since it is contemplated that materials will
be prepared both for printing in the ordinary way and for the
computer input, making the input easily usable may militate to
the disadvantage of the published version. Thus one might restrict
the number of different typefaces to be used, alter the position
of footnotes, or limit the use of tabular or diagrammatic copy.
Still more unfortunate would be any tendency to restrict the
range of terminology used to express the law or the prescription
of particular ways of formulating it.
It should be possible to devise techniques which will both present
the document in the way desired by author and reader to the
extent that these are themselves compatible, while at the same time
making any modification necessary for input to the computer. Since
most of this information will be prepared by some form of direct
entry, it will be necessary to pursue research into the most suitable
form of terminal design, text editing devices and input software,
1974]
LEGAL INFORMATION RETRIEVAL BY COMPUTER
37
none of which is yet in a state suitable for the really sophisticated
operations which will become necessary when a brute machine
replaces an intelligent human being at the initial stage of the
process.
(ii) Searching techniques
Searching techniques are at present largely geared to Boolean
combinations of words occurring in natural language, with some
qualification by reference to separation in numerical and logical
terms. Thus a typical search will ask for references to documents
which contain any one of words A, B or C, followed within the
same sentence by any one of the words D, E or F, but not if any one
of them is itself immediately succeeded in the same sentence by
any one of the words G, H or I. It may well be the case that the
need to cast their thoughts into so rebarbative a form is a prime
cause for the revulsion and difficulty experienced by practising
lawyers in using such systems. This has led a number of workers
to suggest different forms which a computerized search could
take. One which is currently available in some systems, but not
universally and rarely in combination with other forms, makes
use of citations. This is very familiar to lawyers who have been
using citations to find references for centuries. It is true that this
facility is therefore already well provided for by conventional means
in some countries, and notably in the United States. But computer-
ization would make such indices much easier to use both physically
and psychologically: physically because there would then be no
need to turn pages and change volumes at each level of the citation
tracing process; and psychologically because the user could be
certain that all entries were completely up to date. Even here,
however, some changes
in existing systems would be desirable
to permit greater flexibility in the storage, searching and display
of citations than exists at present. For example, second and third
order citations could be retrieved and held in a buffer store for
release upon a first order citation being indicated with a light
pen or cursor, with the other first order citations themselves buffered
and held ready for re-display while this took place. Similarly patterns
of citations could be matched by vectoring techniques in order
to display citations having comparable patterns together.
It would also be possible to mix citations of statutes and cases
in any such system, thus making it possible to collect all the cases
in which a particular statute or part of a statute was cited. These
possibilities raise a more general question in relation to legal
information retrieval. It
is sometimes assumed that retrieval
McGILL LAW JOURNAL
[Vol. 20
operating on the full text of a document is necessarily capable
of conveying all possible information about it. This can easily be
shown to be false since a document’s subsequent history, for
example, is often of vital importance in understanding its signif-
icance and this can clearly never be found in the original text.
Similarly it is often useful to add to the text a large variety of other
pieces of information, some of them informing the user, and others
providing possible parameters within which a search might be
confined. Here too, much more research needs to be conducted
into the nature of this auxiliary information in relation to different
types of legal documents. It
is likely that future systems will
concentrate more and more on the use of terminals and either
multi-access large centralized systems with remote communication
or widely disseminated mini-computers with large secondary stores.
In either case it would be highly advantageous to sub-divide the
data base into smaller units which would normally be searched
independently, though one would have to provide the facility of
combining such sub-units together. This will make for a much
faster and cheaper system, but its introduction must be preceded
by much research in order to establish the optimum pattern for
the sub-division.
The next area in which experiment is needed is that of search
formulation. At present it is quite common to permit only the
use of truncation techniques to assist the user in formulating
his search. He is necessarily left to his own devices so far as
choosing and combining terms is concerned. His most obviously
useful aid is an adequate thesaurus. This will permit the user to
expand his search terms so as to minimize the danger of under-
recall because he has not thought of the synonym or circumlocution
in fact used by a relevant document instead of the term guessed
at by the user in formulating his search. The construction of such
a thesaurus will be by no means an easy matter. It will obviously
be needed to cater for synonyms, antonyms, particularizations and
generalizations. However, all of these notions depend upon context.
In some contexts “will” is a synonym of “stall”, in others of “deed”
and in others of “determination”. “Light” may be an antonym
of “dark” or of “heavy”. In some contexts “shooting” is a particular-
ization of “murder”, in others of “causing an explosion”. In some
contexts “negligence” is a generalization of “overlook”, in others
of “fraud”. It is extremely unlikely that any automatic procedure
alone will ever be able to produce an adequate thesaurus in the
light of these considerations. It will require the very greatest
amount of ingenuity to compile one at all.
1974]
LEGAL INFORMATION RETRIEVAL BY COMPUTER
39
The converse problem is that of providing assistance to the
user to help guard against the danger of under-precision. The most
common device at present is to indicate the number of documents
which will be retrieved on the basis of a particular search formula-
tion. If the number is unacceptably high the user can modify the
search until he has brought it down to acceptable proportions.
This is useful, but by itself inadequate. It would be better if the
documents were put into an order of prospective relevance so that
the user could then examine documents at particular points on
the list to secure some indication of what was going wrong. One way
of doing this, adopted by Professor Lawford in the Quic Law
system”
and by International Business Machines Ltd. in its
STAIRS system, is to allow the user to choose from among a
number of algorithms depending upon permutations of the frequency
of occurrence of a required term in a document and in the total set
of documents, and the number of documents in the total set in
which the required term appears. It is not clear that such algorithms
are necessarily satisfactory, and a scientific study of the efficacy
of these and others in the context of legal information retrieval
would certainly be valuable. A similar result can be achieved
by the mathematically more sophisticated technique of plotting an
n-dimensional vector for each document on the basis of the word
occurrence and importance, and then retrieving documents in terms
of the degree of closeness of match to a similar vector prepared
for the terms of the question.’ 2 It has not, however, been demon-
strated yet that such a system would be applicable to documents
of so varied and complex a character as that exhibited by the
different series of case reports in full text, but such systems are
clearly worthy of further exploration.
One final aid which might well benefit from further consider-
ation is the provision of cueing assistance to the user during
the course of his search. It would be very helpful if he could not
only be given the number of documents which he would retrieve
on the basis of the current search formulation, but also told how
that number would be affected by modification in indicated ways.
Such a system might be programmed to suggest possible deletions
or additions, or the variation of particular operators, qualifiers
or parameters. Further, it would be possible to make more general
11 See Lawford, “QUIC/LAW”, in American Bar Association Automated Legal
Research (1973), 67.
12 For an account of such a system see Vischer, “Das Dokumentationssystem
der Unidata A.G.” in Kybernetik-Datenverarbeitung-Recht Folge 1 (1971), 89.
McGILL LAW JOURNAL
[Vol. 20
advice available for display if the user were to request it, or if
the searcher kept going wrong for reasons which the computer
could recognize, such as repeated over-particularity leading to
under-recall. Here as elsewhere there is much scope for further
research into the sort of assistance which would assist potential
users.
(iii) Testing procedures
they are orientated
There are at present no agreed testing procedures for legal
information retrieval systems. This makes it extremely difficult
to compare results which have been secured by different research
teams in different parts of the world. The two concepts of precision
and recall are by now well established, but they are not entirely
satisfactory, partly because
towards one
particular sort of search, namely that in which information is either
relevant or irrelevant, rather than one in which it is more or less
useful. They depend also upon it being possible to predicate with
certainty of every document in the collection whether or not it is
relevant, and this can usually be true only of relatively small data
bases in relation to which experimentation runs the danger of
being unreliable, statistically unsound or both. Neither are there
agreed standards for counting results. Thus if one researcher finds
all nine out of nine relevant cases in four searches, and one out
of four in the fifth he might aggregate his figures and state that
his recall is 90%, while another might calculate each search
independently and then average out his recall as 85%. This sort
of consideration is usually left unexplained in the presentation
of experimental results.
In some obvious respects in which systems (especially those
intended for commercial operation) should be tested, such as the
time and expense of searching, there seems to have been no con-
trolled testing at all. It is true that these factors may be difficult to
measure, but that is no excuse for abstaining from any attempt
to do so.
(b) Consequences
What difference would it make if all these steps were taken
and a satisfactory system of legal information retrieval by computer
available? Some might say that the legal system would collapse,
choked to death by authority, but the concept of a satisfactory
system is intended to guard against this riposte. In fact the
changes are quite unimaginable in toto, and all that is possible
1974]
LEGAL INFORMATION RETRIEVAL BY COMPUTER
41
here is a very brief indication of the possible impact upon legal
education and the organization of the legal profession with some
international implications.
Legal education would clearly have to change to accommodate
the new technology. This does not mean that lawyers would have
to understand computers; they would merely have to understand
how to use them. It is also unlikely that a new system will be
introduced for legal information retrieval before computers become
much more a part of everyday life than they are today. This will
probably mean that the basic principles of the computer, including
elementary programming and keyboard operation, will be taught
regularly in schools. This assumes that voice input will not by then
have developed sufficiently
to have made keyboard operation
obsolete. It is quite possible that the necessity for direct keyboard
entry by the legal practitioner is one of the things inhibiting the
spread of conversational systems at present. Thus it is likely that
in the future prospective lawyers will already be equipped with
the technical and manual skills necessary for effective use of
computerized systems. This will
leave the law schools of the
future free to concentrate upon the specifically legal side of such
systems, which is just as it should be and is not at present. It is
further envisaged that law schools will not need to establish
special courses in the use of both systems any more than at present
they have’special courses in the use of indices, books and libraries.
Rather, the impact of the new technology will influence the teaching
of substantive law. Students will find their attention directed much
more towards the ways in which rules and concepts are expressed,
and will approach meaning much more from that point of view
than they do at present. For those brought up in a tradition of the
importance of the analysis of usage this will seem no bad thing.
It is also, of course, likely that a secondary change will take place
in the usage itself under the pressure of such close attention. Such
emphasis will inevitably spill over into the world of legal publishing.
The student book market will have to change to accommodate the
greater stress on modes of expression, and new sorts of books
suggesting possible search patterns will appear, although most of
the change will lie in the incorporation of this change of emphasis
into existing publications.
Another factor which probably plays a large part in inhibiting
the acceptance of systems at present is that of cost. Quite apart
from the cost of acquiring the data base there are expenses associated
with securing access to a computer and with using it once access
has been obtained. The legal profession is not ideally organized
McGILL LAW JOURNAL
[Vol. 20
to absorb these costs since it is mainly split into small units. 3 This
means either that firms must subscribe to a general service, combine
together into larger units for such a purpose or avail themselves
of the mini-computers which are becoming increasingly common. 4
The other factor in the current structure of the legal profession
which may be inimical to the development of retrieval services
by computer is the distinction between solicitors and bar in many
jurisdictions, either de jure or de facto. Where this distinction
exists it exacerbates the division between the different types of
search which may be required, and in such jurisdictions it is most
unlikely that the same service would satisfy both branches of the
profession. Similar considerations also apply to other segments
of the legal profession, such as those in government (local and
private), industry and the universities.
Finally something should be said of international implications.
As the number of supra-national bodies multiplies, as treaties
between states increase and as private citizens increasingly trade
and travel in foreign countries, so the demand for knowledge of
other legal systems will increase. It is unlikely that any system
designed only to retrieve foreign or international law will be viable
except perhaps in a few highly specialized areas. The most that
can be done therefore is to ensure that so far as possible data
prepared for national systems is compatible with that prepared for
other systems. This requirement should not in any way inhibit the
development of national systems, but there are many features of
a computer system which are entirely arbitrary in the sense that
they can be accomplished as easily one way as another. In such
cases there is an obvious advantage in as many different systems
as possible doing the same. Some work is already being done on
this problem in Europe under the auspices of the Council of Europe,
which in 1969 set up a committee to consider the harmonization of
legal computer systems in member states. It would be helpful if
this sort of approach could be widened, perhaps as a result of
an initiative by an agency of the United Nations.
SUMMARY
This article has attempted to describe the progress made in legal
information by computer during the past fifteen years, and to
Is For the precise distribution in Canada see Operation Compulex, Depart-
ment of Justice, Ottawa (1972), app. 1, graph C.
14 See Hoffman, Survey of Law Firm Computer Use –
trics Journal 42, 86 for an account of these tendencies in the United States.
1971, (1971) Jurime-
1974J
LEGAL INFORMATION RETRIEVAL BY COMPUTER
43
show how the pattern of that growth was largely determined by
decisions made at the University of Pittsburgh in response to a
rather special set of problems. It has been suggested that many
of the difficulties
in commercial application currently being
experienced stem from failure to look hard or radically enough at
the nature of legal searching. Some attempt has also been made to
indicate directions along which thought might now be channelled
in the hope of realising the benefits which computers may bestow
upon the law.
