MULTILEVEL DESCRIPTION, MULTILEVEL INHERITANCE, RELATIONS/LINKS:
CONTENT AND CARRIER
by
Mary Lynette Larsgaard
Alexandria Digital Library/Davidson Library
University of California, Santa Barbara
Version 0.6
September 22, 1996
Introduction
"The major thrust of cataloging should therefore be in building conceptual models of works and their agents, by which people usually try to find such works .... in principle, MARC is independent of AACR2 and of any other set of cataloging rules."
Heaney, Michael. 1995. Object-oriented cataloging.
Information technology and libraries (9/95):15-53.
"0.24. It is a cardinal principle of the use of part I that the description of a physical item should be based in the first instance on the chapter dealing with the class of materials to which that item belongs. .... In short, the starting point for description is the physical form of the item in hand, not the original or any previous form in which the work has been published."
Anglo-American cataloguing rules. 1988. 2d ed. rev.
Chicago: American Library Association.
These two quotations form the basis of the following
paper, which agrees completely with the first and disagrees equally
with the second, taking instead the stance that intellectual content
is of primary interest to the user and physical form important
but secondary.
The Relationships
Call them what you will - cartographic materials
or spatial data - these items about 80 percent of the time have
some sort of link or relationship with others of their ilk. At
the same time, what was allowed by AACR1 and by USMARC to show
the nature of these links was certainly thorough but unfortunately
very time-consuming. Then, with AACR2, effective January 1, 1981,
multilevel description for parent-child relationships became an
option, which by implication allowed other relationships to be
expressed within the bibliographic record.
Unfortunately, USMARC - not only the best but the
most heavily used machine-readable format for bibliographic description
in U.S. libraries - did not. As has been generally true in the
last fifty years for U.S. libraries, AACR proposes, the Library
of Congress (LC) - and by extension USMARC - disposes. That is,
when it comes to a matter of the current edition of AACR stating
a rule, and LC stating that it will follow a different practice,
many libraries unswervingly follow LC, for many sheerly practical
reasons having to do with the economics of accepting existing
LC, and LC-following libraries', copy as is. USMARC is intended
for use as a communications format for bibliographic description,
but not for any one given set of cataloging rules, although for
historical reasons, AACR in its various editions is very fully
provided for.
USMARC did not provide for linking relationships
as specified in AACR2 until relatively recently. There are many
linking relationships possible, and they are often in reciprocal,
paired fields. For the purposes of this paper, the links downward,
from what might generally be called parent to what might generally
be called child, will not be discussed in this paper, but rather
only the links from child to parent. This paper is an outcome
of the cataloging done for the Alexandria Digital Library, during
the initial stages of which a database computer engineer - after
inquiring of me how often users would need to know the names,
etc., of all sheets belonging to a series (e.g., all 57,000 sheets
in the U.S. Geological Survey 1:24,000-scale series), and hearing
my response, "Seldom" - stated that the reciprocal fields
were not needed, that for the few times such information were
requested, software could search all linking-field $w's in child
records (control number for e.g., map-series parent, 772$w) for
those containing the given control number.
The fields upon which this paper is focused are as
follows:
772 link to parent from child
773 link to host from part
775 link to prior edition from newer edition
[Query: does one link always to the first edition
but not between different editions? or are all editions linked
to each other? And are multiple links possible - that is, in the
case of a new edition of a U.S. Geological Survey topographic
quadrangle, from one edition to the earlier editions, and also
to the parent record? oR are links in chains - that is, newest
edition to earlier editions, and from earliest edition to parent?]
776 link to "original" physical form from other forms
[Query: same query as for 775]
787 nonspecific relationship
These fields are of especial interest to catalogers
who want to give users the best possible access but have no interest
in the reinputting of information which classic methods of cataloging
require. We thus have the interesting situation of USMARC making
possible cataloging practices which, except for parent-child,
are not specifically mentioned in AACR (latest edition).
Following is a table giving the AACR2R method of
making clear a relationship, with the USMARC method(s) in the
second column.
I. Parent/child
Unfortunately, USMARC lumps together some very different
relationships together under 772, including supplement and whole/part.
For the purposes of this discussion, the focus is on the parent/child
relationship, which is an excellent expression of the most frequent
of relationships within the spatial-data world, that of map series
(series as a whole/individual sheet) and of remote-sensing images
(e.g., air-photo flights; satellite-image missions).
AACR2R USMARC
I. Parent/child
i. series 4xx, 8xx
ii. multilevel description 772
II. Host/part
i. "in" analytics [fill in these fields]
ii. multilevel description 773
III. Other edition
i. Make full record. 250
ii. multilevel description 775
IV. Other physical form
i. Make full record. 533 (details of reproduction); 534
(details of original)
ii. multilevel description 776
V. Other unspecified relationship/complex relationship
i. Make full record. [e.g.?] 580
ii. multilevel description 787
Several questions immediately come to mind here about linkages:
a. are there some situations - such as large-scale topographic series - where one would use 772, and others - such as monographic series, where the child is much more loosely tied to the parent, with far fewer fields in common from child to child, in fact only the author/title of the parent - where using 4xx/8xx is the best procedure?
b. how many linkages does one make in such situations as map series, where many different editions of a single sheet exist? are the links from more current editions back to the first edition, but not to the parent record? are links to parent record and to previous editions? and so on. (1)
c. similarly, when a map sheet is scanned into digital
form, is the link from the record for the scanned item to the
record of the hard-copy original? is there also a link to the
parent record for the map series?
As was briefly indicated above, the parent/child relationship is especially important for spatial data, since the vast majority of all materials participate in that sort of relationship; for example, such items as:
a. map series; e.g., U.S. Geological Survey 1:24,000-scale topographic sheets; a total of about 57,000 sheets;
b. air-photo flights; aerial photographs most frequently are part of flights, with ca. 200-300 frames per roll.
c. satellite images; e.g., the first Landsat satellite
was launched in 1972, so by now there are literally millions of
images
and most recently:
d. geographic information systems: 2 different types of relationships:
i. tiles (analogous to sheets in a map series)
ii. layers of thematic information that cover the
entire area; each thematic layer will cover (or should cover)
the entire area, rather as in a national atlas - each sheet covers
the same geographic area and shows a different subject. There
is rarely much common metadata between coverages, since each generally
has different lineage. It would also be difficult to devise a
concept that survives from one GIS software to another. It will
be interesting to see how this system works when one applies it
to digital data, and specifically to vector data.
Approaches
What is needed in USMARC is a multilevel-description
field, that expresses what role or roles any given record has.
There are several ways to do this:
1. the national map collection of Canada, which is part of the National Archives of Canada, uses (in GEAC) the following scheme:
001 parent record
002 subrecord
003 both parent and subrecord
null monographic item
(Parker, Velma. 1990. Multilevel cataloging/description for cartographic materials. Western Association of Map Libraries information bulletin 21:86-96.)
2. the Alexandria Digital Library currently uses:
root; node; leaf
A change that is being considered is to use numbers:
- either: monographic item null
parent 0
child 1-n
- or: 0 not compound
1 parent
2 child at second level
3 child at third level
(and so on)
Currently a document dealing with this matter generally
is up for review:
Functional requirements for bibliographic records,
draft report for world-wide review. May 1996. Frankfurt am Main
: Deutsche Bibliothek [for] IFLA Universal Bibliographic Control
and International MARC Programme.
This report - which is recommended by the IFLA Study
Group on the Functional Requirements for Bibliographic Records
- has as its base the idea that there are three tiers involved:
1. the work;
2. the manifestation; and
3. the copy.
AACR in its various editions has "developed
in a world where a document was relatively stable over time ..."
(Dorman, David. June 7, 1996. Re: edition statements. Saratoga
Springs, NY: Skidmore College. email message ID 31B88E99.430B@skidmore.edu).
The IFLA report attempts to deal with a world where that is no
longer true for many "documents."
Host/part is perhaps the least common of these relationships in the spatial-data arena, but it is nonetheless much needed. For many years, map librarians have needed quick access to maps bound in other publications, e.g.:
a. individual plates in atlases [cite Nancy Edstrom etc. on tables of contents for atlases]
b. maps in periodicals: The American Geographical
Society's index to maps in periodicals is certainly extremely
useful.
The other-editions situation very frequently occurs
in topographic series, and to a far lesser extent with some monographic
maps, such as the geologic map of Africa issued in different editions
by Unesco, or the maps of states at 1:500,000 issued by the U.S.
Geological Survey.[get citations]
There has already been considerable discussion about
cataloging an item that appears in several different physical
forms. What is needed here is a briefer way to catalog, I would
suggest using the 776 fields, to give us what Barbara Tillett
calls a machine version of the old printed-card dashed-on entry.
[cite Barbara's Dublin-core-meeting paper]
What happened in these dashed-on entries was this.
Let us say, for example, that we have a topographic map series
of [use one of those AMS map series records], with an index. The
catalog record would look like this:
United States. Army Map Service.
...
--- Index. ...
The "dashed-on" part is an analog of the
bibliography in which one has one author with more than one publication
listed; in this case, the author's name is shown in all entries
except the first as a brief underlining. What makes this such
an important option for catalogers is that one may have one intellectual
entity in several different forms - e.g., monographic book; CD;
VCR tape. It is not useful to library users to offer them lengthy
sets of citations for what is the same intellectual content on
different carriers. This is especially important given that most
users are not interested in looking at any more than about 35
citations. The problem here is that AACR2R states in 0.24 that
an item is to be cataloged first of all by working with the chapter
dealing with the carrier, before intellectual content is considered.
This is not in accord with user requirements. It is extremely
rare for a user say, "I need a CD and I don't care what's
on it." Certainly carrier is important to the user, but it
is secondary. Content comes first.
Other physical form is increasingly important for
spatial data with the continually growing presence of spatial
data in digital form. There are a couple of different approaches
that one might take here:
a. describe a number of different physical formats in the same record;
b. describe the different physical formats locally;, in e.g. NOTIS holdings (HLD) screens;
c. describe each separately and fully;
d. describe each separately but by using linking
fields do not repeat any information in a child that appears in
a parent.
In the past, microform was the chief situation of
this type, with some 35mm-slide, facsimile and photocopy occurrences.
More and more users say to the map librarian, "I need x spatial
data in digital form." The most recent major event in this
area in the United States is DRGs, Digital Raster Graphics; these
are scanned versions of the several map series in the U.S., the
1:24,000-scale, 7.5' topographic quadrangles, the 1:100,000-scale
sheets and the 1:250,000-scale sheets. And how are we to deal
with the matter of one digital file that is available in different
file formats, e.g., GIFF and JPEG?
Other unspecified relationship [so far, I haven't
come up with one of these for cartographic material - suggestions?]
Getting A Bit More Specific
In Appendix A are given several different kinds of examples:
a. map from a volume;
b. different editions of one USGS 1:24,000-scale sheet;
c. the same spatial-data item cataloged first according
to AACR2R and second using the alternate method that USMARC makes
possible.
All are USMARC-tagged; as a part of my death-to-all-coded
fields initiative, USMARC tags 1xx-8xx are emphasized. [Possibility
for future: Scans of portions of some of the maps, to illustrate
from whence the "bibliographic" information came.]
The USMARC-method makes for observably shorter records.
Note especially in these records that for the USMARC records only
the $w (control number of record being pointed to) is used; putting
in all fields is extra, repetitious work for the cataloger.
It is essential to keep in mind that these USMARC-method
records are ONLY what the cataloger inputs. What the general user
sees is another matter altogether, and a difficult question it
is, to figure out how we are going to display this complex situation
to the user in a clear, unambiguous way. The three obvious options
are:
a. parent
- child
(this is like the old dashed-on method)
b. child
- parent
c. child record that is an agglomeration of parent and child fields, and that looks almost exactly like the record of the AACR2R version. See example __ in Appendix A of a record for a sheet of the 1:24,000-scale USGS map series.
[see appendix for examples]
Parkinson's Law and the general cussedness of life
makes one suspect that the favorite method of users will be whatever
takes computer engineers longest to code. Another look is enough
to convince or at least strongly to persuade us that the way to
do this so that the cataloger does minimal inputting is to use
templates, as for example happens when a cataloger working on
OCLC, faced with cataloging a new edition of title, calls the
previous edition, types, "new" at the command line,
and works with a record that has 1xx-8xx intact, with only the
001-099 fields blanked out. This will probably have the same general
effect as the USMARC-method examples in Appendix A without computer
engineers having to write code to meld without overwriting fields
a child and parent record to the example given previously.
Yet another possibility is not to have any note fields
in the child records, and instead to have any necessary notes
(especially noting variances, e.g., "Phrases at head of title
differ") in the parent record. In any case, it will be important
to use explicit field labels (not in libraryese) noting roles
and relationships, with, with the default display being minimal-level/brief
cataloging fields. Also, the different relationships may need
to be shown in different ways.
The overwhelming importance to cartographic materials
of being able to link and show relationships is depicted in the
following worst possible case (see Appendix A for records):
a. parent record - all USGS topographic series
b. parent record - 1:24K USGS topographic map series
c. parent record - orthophotoquada
d. individual quadrangle: Santa Barbara 1952
e. individual quadrangle: SB 1967
f. individual quadrangle: SB 1988
g. individual quadrangle: SB orthophotoquad 1976
h. individual quadrangle: SB orthophotoquad 1980
i. ??Digital Elevation Model for SB 1:24,000??
j. ??DLG for 1:100K??
Let's start out with relatively easy matters first,
and then link together the entire set.
Host/part
PART RECORD
This item is physically a part of: HOST RECORD
Note: this does not link with the rest.
Source
DIGITAL ELEVATION MODEL
Source/lineage: melded record
Parent/child
MELDED RECORD
or
PARENT
- CHILD
or
This CHILD RECORD ("in" analytic approach)
is a part of: PARENT RECORD
Other editions
FIRST EDITION
Other edition: RECORD
Other edition: RECORD
Other edition: RECORD
etc.
Other physical forms
HARDCOPY TOPOGRAPHIC SHEET
Scanned into raster digital form: RECORD
file extension: .gif
file extension: .tiff
file extension: .jpeg
Microform: RECORD
Facsimile: RECORD
All together:
Parent
- child
- other physical form:
- facsimile
- microform
- scanned
- gif
- jpeg
- etc.
- other edition
- as above
- other edition
Child as source
- link to Digital Elevation Model
- link to Digital Line Graph
Ending Thought
We have two major goals in mind:
a. present simply, and clearly, complex relationships (e.g., multiple versions) to the users but only as much information as they need;
b. simplify the creation of catalog records by catalogers
(Martin, Giles. December 17, 1995. Re: field 856
considerations. New South Wales: University of Newcastle, Quality
Control Section, University of Newcastle Libraries. Message ID
Pine.3.89.9512181017.A6847-0100000@dewey.newcastle.edu.au)
FOOTNOTES
(1) The Library of Congress is exploring several
different methods that have been used at LC for whole/part and
parent/child relationships, to try to standardize for maps, mss,
photos, etc. Kay Giles is in charge of this effort.