The Australian Library Journal

The Australian National Bibliographic Database and the Functional Requirements for the Bibliographic Database (FRBR)

Bemal Rajapatirana and Roxanne Missingham

The development of the conceptual 'Functional Requirements for Bibliographic Records' (FRBR) Model enables records to be considered in terms of contextual relationships. Developments in software can capitalise on this to significantly improve the display of works through surfacing of these relationships. This paper reports on an investigation of the Australian National Bibliographic Database to understand the implications and potential issues of applying FRBR in this environment.

Manuscript received September 2004

This case study was presented in February 2004 to a seminar on FRBR run by the Australian Committee on Cataloguing

Library catalogues and union catalogues have, for many decades, been based upon records stored in MARC and AACR2 format. While this has given stability and consistency, the emergence of a new concept, the Functional Requirements for the Bibliographic Record (FRBR), has enabled a new light to be shone on management of bibliographic records. The FRBR conceptual model proposes descriptions of library materials with four levels of description:

• A work (which is realised through)
• An expression (which is established in)
• A manifestation (which is exemplified by)
• An item.

In posing these relationships the model offers the potential to move the intellectual activities undertaken by cataloguers into a new environment where information is not stored in a 'flat' record but one where relationships enable searching and display based on a more flexible understanding of resources.

FRBR can enable users to find resources which are related through catalogue entries which, while containing the information traditionally described in bibliographic records, are 'clustered' together in a new way. While it is relatively early days for FRBR, the RLG and OCLC have applied these concepts in new services for users - the Red Light Green catalogue aimed at undergraduates and OCLC's Fiction Finder.

Australia's National Bibliographic Database (ANBD) is a union catalogue of over 37 500 000 items held in Australian libraries and a national bibliographic database of resources including Australian online publications. The Kinetica service enables Australian libraries, and individuals, through their libraries, to access the ANBD and other databases. It also supports shared cataloguing and sharing of resources. The National Library has supported national discovery services online for over twenty three years. The service fits within the Library goal to:

provide rapid and easy access to the wealth of information resources that reside in libraries and other cultural institutions and to break down the barriers that work against this. (National Library of Australia, 2003)

A project was undertaken in 2003 to investigate the potential to support and exploit FRBR concepts for the ANBD. The scope of the inquiry was:

• analysis of records on the ANBD to determine those records which would be clustered through FRBR;
• in depth analysis of a popular Australian work; and
• consideration of issues relevant to the application of FRBR to the ANBD.

Methodology

To investigate the records on the ANBD which were related under the FRBR model two tests were devised. The Library of Congress (LC) FRBR Display Tool was used for these tests. The OCLC algorithm was not used as it was not available at the time in a form that could be used to test a data file in this manner.

The first test was to identify the 100 most commonly occurring uniform and name-title index entries, in essence the 100 most commonly occurring work headings and then, extract the linked bibliographic records and feed these through the LC display tool. Once this was done, analysis occurred to extrapolate the number of records (or manifestations) per work, qualify missed items and identify any material type trends in the result set. The data collection included the complete set of bibliographic records pertaining to all works as well as sample subsets of records relating to single works. They were then processed with the LC display tool in order to compare the affects on and across specific genres.

The records were analysed to identify the most commonly occurring Australian name titles and title headings. The authority files were used for this analysis as the most reliable source for this study. Authority records marked as 'Australian' authorities were linked to their corresponding index entries, which were then used to identify the bibliographic records for Australian works.

The second test was to expand this process to a larger result set - the entire ANBD in order to identify the top 1000 results as a means of comparing ANBD 'work to record' ratios against those defined for WorldCat (1.5 manifestations to work) and the RLG Union Catalog (estimated to be 3 manifestations to a work). The second test was also intended to apply a different algorithm, if made available by OCLC, and deploy this on a Unicode-enabled platform to assess the impact of including CJK records. Owing to the limitations on the availability of a fully encoded alternate algorithm and a Unicode platform, the second test did not proceed.

In additional analysis was undertaken into the display of records using the LC FRBR Display Tool and an in-depth analysis of a single Australian work.

Common works study

Common works, being works for which more than one expression or manifestation occurred in the ANBD, were identified using name and uniform title index headings with the most number of bibliographic records attached. The MARC fields used to identify these work headings consisted of tags 130, 240, 243, 730 for Uniform titles, and any tag 100, 110, 111, 700, 710, 711 with a $t subfield for Name/Titles.

Initial analysis

The first listing of commonly occurring work headings was problematic as a number of inappropriate headings ranked in the top 100. A common problem was the impact of series or other headings established as collocating devices within the ANBD title indexes. Though works in series are legitimate entities within the FRBR conceptual framework, they generally represent work-to-work relationships that were not a primary consideration in the NBD analysis.

This first list was refined by applying global exclusion criteria, that is, automated SQL scripts that found common conditions on which to exclude a heading. This resulted in a more accurate and meaningful name/titles list but was less successful with uniform titles. The following exclusion criteria were applied to name/title and uniform title headings.

Both lists had to be further refined by manually excluding specific name/titles headings prior to extracting a dataset. While manual compilation was possible with a limited set of work headings in this test, this type of scoping would be difficult to perform in a full scale production database environment. For example, 'United Nations Association of Australia. Media peace awards' was manually de-selected as it failed to meet the global exclusion criteria and there was no way to apply criteria that would reliably eliminate just this heading or those of a similar nature.

In generating this list of commonly occurring work headings the presence of duplicate headings was noted, but in most instances this did not affect the ranking of these headings or the aggregations resulting from the LC display tool. The FRBR display tool ignored this duplication and a future option could include normalising and aligning headings against an authority file prior to the identification of works. In a few instances duplicate headings did affect ranking with the numbers of bibliographic records relatively evenly split between two identical or nearly identical terms, for example near-identical headings for 'Bible. English. Authorised' and 'Bible. English. Authorized' had bibliographic records equally dispersed between the two headings as did identical duplicate headings for 'Chopin. Piano music. Selections' and 'Liszt. Piano music. Selections'.

A cataloguer's decision to enter a work under either a title or name/title heading affects both the ranking of work headings as well as the clustering of bibliographic records. For example, the heading 'Aesop's fables' has significantly more items attached than the headings 'Aesop. Aesop's Fables' or 'Aesop. Fables'. This is problematic in that the identification of common works and the algorithms used for clustering do not recognise these headings as related or cross index boundaries.

Final analysis of common works

The final list of most common work headings resulted in 20 856 linked bibliographic records extracted from the ANBD. These bibliographic records represented manifestation level metadata. The data set extracted against name-title work headings yielded 13 561 manifestation level records and 4490 works were identified. The data set extracted against uniform title work headings yielded 7025 manifestation level records with 2408 works identified.

The final list of most common name-title and title headings resulted in 13 561 and 7025 bibliographic records extracted from the ANBD. In terms of work to manifestation/record ratios, this represents a 1:3 ratio for works identified through name-titles and a 1:2.9 ratio for works identified through uniform titles. Collectively, this represents a work to manifestation/record ratio of 1:3. Both RLG and OCLC have studied this ratios for their catalogues and discovered ratios of three records per work and 1.5 manifestations per work respectively. In comparison, records in the ANBD are more like those in RLG's union catalogue than OCLC's WorldCat.

The composition of both lists, after the application of the global exclusion criteria and manual de-selection, was quite consistent and largely anticipated. The lists reflected literary, religious and musical works. The following tables display the ten most commonly occurring work headings for each data set extracted from the ANBD.

A comparison of the top ten works on the ANBD with similar rankings from OCLC's World Cat (Hickey, 2002) indicates that there is much in common between the two resource sets. The following lists demonstrate this even though the OCLC work-set identification was also performed against segments of their WorldCat database (representing different collection types) and the rankings for ANBD have not been combined.

Display of records in FRBR

An analysis of the potential of display ANBD records in FRBR was undertaken. The Library of Congress FRBR Display Tool, version 1.1 became available in 2003, based on analysis done by the MARC Standards Office which culminated in the publication, Displays for Multiple Versions form MARC 21 and FRBR. The tool is a program that converts MARC bibliographic data into hierarchical FRBR displays that are grouped and sorted by the work, expression and manifestation entities. The observations that follow in no way diminish the value of the tool for analysing the FRBR potential of the ANBD data sets.

The tool performs an XML conversion which could be problematic for some implementations where locally defined MARC tags must be incorporated into standard XML DTD. The ANBD uses almost no local tags but may need to consider adapting an existing DTD (such as the MARC XML DTD) to accommodate data from external sources (overseas and local) and possibly also support obsolete data within the ANBD if superseded MARC elements are not defined in the current MARC XML DTD.

The tool is very sensitive to data errors within ANBD records particularly in the transformation from MARC to XML for example, subfield delimiters (1F) or punctuation (CCA6) occurring mid-field or in coded data fields caused the tool to abend. It would be preferable for records that failed validation to be written to an error file and the tool continue processing the remaining records. Other common problems could benefit from automatic substitution of default values.

There were some doubts as to whether the tool would scale to accommodate a data set the size of the NBD (approx 13 million records). It was not possible to test this given the technical and resource constraints in this investigation. A UNIX environment may be required to clean up files that fail validation as a Hex editor and standalone PC will have file size and memory limits.

The tool does not generate statistics upon completion of processing. It was possible to gather statistics employing functionality within an XML editor but this involved a second process. The XML editor provides an occurrence count against each MODS element in the file. Even here the statistics for the <manifestation> element had to be interpreted carefully as they do not reflect actual manifestations due to the syntax structure. The <manifestation> element is a parent element or wrapper for each <imprint> element with the latter representing the actual number of manifestations entities.

The sorting rules imposed by the tool lead to curious and sometimes misleading displays. For example:

• Works with Author main entry filed alphabetically but before any works with title entry. This presented problems where cataloguers had made different choices of main entry.
• Expressions carry text labels but are sorted alphabetically by the code value in LDR/06 so 'text' (LDR06=a) files before 'sound recording' (LDR06=j), and 'sound recording' files before 'software, multimedia' (LDR06=m).
• Manifestations were sorted chronologically by date descending under each Expression. Where manifestations shared the same date no other sort rule was applied.
• Some leader codes were represented as their code values and this was possibly a display conversion error, for example, Form: LDR6=o

The algorithm within the tool appears to generate identical but duplicate works, possibly due to a fault or as yet misunderstood feature of the algorithm. The explanatory text accompanying the LC Display Tool algorithm does not reveal any clues to interpret this behaviour. The examples below should have generated two manifestations under 1 work if the match was only performed on a 1XX and 240 fields, but instead it generated 2 works with 1 manifestation each:

The algorithm does not differentiate between musical and non-musical sound recordings for display purposes. Both forms are assigned the same expression label 'sound recording' leaving the user to determine which is applicable, presumably from the physical description area but this can be inconclusive.

The Library of Congress tool is a very significant step forward, simplifying the display of larger results sets to some extent. While the tool does aggregate a large work set retrieval into more manageable and labelled expression subsets, where these contain numerous manifestations there is still the sense of wading through a list of results. It omits fairly vital information in that there is no display of summary holdings. However it does support links into a catalogue search (in this case the LC catalogue). However, by coupling the logic within the algorithm with a more interactive, linked and layered display interface, a user's navigation from work through to items could be a better one. This research suggests that an FRBR display tool could be applied to the ANBD, which would need to be developed beyond the current LC tool.

Clustering and display algorithms

The two algorithms which were chosen to further assess clustering and display issues were the LC FRBR Display Tool algorithm and the OCLC Work-Set Algorithm. The most noticeable difference is that the LC FRBR Display Tool deals with a broader range of entities. It seeks to identify, cluster and display works, expressions and manifestations whereas the OCLC Work-Set algorithm is only designed to identify a work-set and does not address display issues.

The other key differentiator is that the OCLC algorithm exploits data in authority files. It does so by mapping variant forms of names in the bibliographic record to the form of name authorised in the authority file according to certain rules, for example where exact match on name fails, for multiple matches attach to the form of name used most frequently (approximates to the form with the most number of bibliographic records attached), or ignore birth and death dates to get a variant form to match. This makes the OCLC algorithm more reliable for the identification and clustering of works.

There are many similarities in the data elements that both algorithms exploit when constructing author and title parts for the work keys. For example, both use the same subfield codes for the construction of authors; though there is less overlap in the construction of the title (generally fewer subfields are present in the OCLC algorithm).

Similar data correction and normalisation is employed in both algorithms. For example in constructing the title key both ignore non-filing characters (or leadings articles), remove white spaces (OCLC removes leading or trailing blanks) and other punctuation. Again there are some differences in how titles are normalised for example, OCLC removes authors' names from the beginnings of titles. OCLC indicates that work-set keys are normalised according to the NACO rules to simplify them prior to matching.

Neither the LC or OCLC algorithms address how to merge duplicate manifestations (either in the data or on display). This is something that would be quite desirable in a large bibliographic system. For the ANBD a solution will therefore need to be developed using the results of this analysis.

In-depth study: Waltzing Matilda

Waltzing Matilda, a very popular Australian song, was selected for in-depth analysis as a means of investigating the possible impact of FRBR on a set of records of an Australian work or works. Waltzing Matilda provides an interesting challenge as it:

• is a work which can be actually considered to be a number of works - while A B (Andrew Barton) Paterson has been credited as the author of the popular form of the work written while staying with the Macpherson family on a property in north-west Queensland, it is now thought to have mostly been written by Christine Macpherson,
• Christine Macpherson acknowledges she adapted the tune from an existing folk song,
• the work has been realised as a poem, song, performances, and a movie,
• there are also regional variations including Cloncurry, Buderim and Queensland versions,
• many performers have produced difference versions including O'Keefe and Nathan,
• it was the highest ranked or most common Australian work on the ANBD.

The in-depth study was conducted by extracting records with authorised name headings 'Cowen, Marie. Waltzing Matilda' and 'Paterson, A B (Andrew Barton), 1864-1941. Waltzing Matilda'. In total 46 works, 55 expressions and 154 manifestations were identified. The three most prominent works appeared under the following headings:

Cowan, Marie. Waltzing Matilda (150 manifestations)

Paterson A B (Andrew Barton), 1864-1941. Waltzing Matilda (35 manifestations)

Waltzing Matilda (7 manifestations)

The analysis revealed a number of significant issues in applying FRBR principles to these records:

• comprehensiveness of records set: the initial search on the two authorised name - title index entries did not provide sufficient records. A number of keyword searches were undertaken to obtain the maximum number of relevant records.
• confusion about the number of works: regional variations, such as the Queensland version, did not have a heading or authority work. Information suggests that the Queensland version has the same lyrics but a different tune. It is difficult, therefore, to establish whether these can stand as separate works.
• in consistent or incomplete data in records, such as the absence of language codes or records with an added entry for the name of an arranger and information in the 245 field indicating the representation was an arrangement, but without recording the creator role of 'Arranger'
• use of notes field to record information preventing automated identification of details for FRBR presentation, for example:

  Author: Not entered under author

  Work: Material relating to Waltzing Matilda 1905-1969

  Form: text - English

  Edition:

  Title: Material relating to Waltzing Matilda

  Imprint: 1905-1969

  Physical Description: 2 items (6 p.)

This record provides an indication of Nathan's role and that the expression is based on the work by Macpherson only in the notes fields

• tensions exist between literary and musical works of Waltzing Matilda, as it is difficult to definitively determine whether the literary work is expressed through a musical representation or the reverse.
• clustering by performance medium would be another useful approach. A small sampling exercise indicated that seventy-five per cent of records (sample of twenty piano works, six orchestral works, and ten vocal and choral works) carried codes in the 048 field (Number of Musical Instruments or Voices). Where the code was not evident, it appeared possible to massage the data to supply the code for the type of instrument but not always the number of instruments.

From this study it was concluded that the application of FRBR to a work or related works with a wide variety of expressions and manifestations may require some work with data in addition to complex rules for display.

Feasibility of FRBR for the NBD

The research undertaken enabled the Library to conclude that it would be possible to apply FRBR to the ANBD. The LC Display tool could be applied, with some further work. While overall quality of data was an issue, several common problems which could be addressed programmatically were identified:

• material type: incorrect coding will create new expressions incorrectly. Advice should be issued to cataloguers to control for this problem.
• incorrect use of name/title: for example headings 'Aesop. Aesop's Fables' and 'Aesop. Fables', should refer to 'Aesop's Fables'
• incorrect main entries: for example, incorrect attribution of main entry to a performer can also lead to the generation of duplicate works for example in W1 and W2 are actual duplicate manifestations but each appears under a separate work heading:

  W1	W2
  Author: Bartoli, Cecilia	Author: Not entered under author
  Work: Italian songs	Work: Italian songs, canzoni
  Manifestation Title: Italian songs: Canzoni	Manifestation Title: Italian songs: Canzoni

• incomplete records, for example, records that lacked codes, particularly language codes were set out as near duplicate expressions for example two expressions were established under the work 'Cowan, Marie. Waltzing Matilda'. One expression was for 'Form: notated music' and the other was for 'Form: notated music - English'. The only reason for this demarcation is that one set of bibliographic records carries a language code and another does not.
• inaccurate records: for example, 'Alice's cookbook' or 'Alice's cook book' will produce variant works, a result of simple data quality issues.

There is also a high correlation between expression and work headings, which is indicative of two main problems. One is a possible interpretation of the cataloguing code as set out in the Anglo-American Cataloging Rules, 2nd ed (AACR2) and the other is the extent to which the cataloguer will choose to record those data elements that pertain to modes of expression. The boundaries between work and expression level cataloguing are not yet adequately defined in AACR2 and unless the need to delineate expression level elements is understood and cataloguing interfaces are more adequately set up to support their input, bibliographic data in MARC records will not expose all expressions.

Many of the data issues listed above will not be easily resolved. Problems such as coding errors and duplicate index entries have been identified as issues for the NBD to be addressed by a program of quality improvement work. Some problems can also be moderated by adopting a different algorithm (for example, one that utilises normalisation of the data against the authority file).

Despite the impact that data quality issues can have on FRBR, the outcomes suggest that it is worth implementing FRBR, if only selectively on the NBD. There is some evidence to suggest that the application of FRBR is most useful to that subset of bibliographic data for popular literary, musical and theological works. These are inevitably works with the most number of bibliographic records and as with most works of the imagination are generally subject to a higher number of variant interpretations and therefore expressions.

However, as most of the algorithms appear to have a neutral affect on works that fall outside these categories, there are few reasons to limit the algorithm to just this portion of the NBD. Restricting the subset to which FRBR is applied may offer some savings in terms of staff resources and improvements in terms of the quality of the FRBR database as less time and effort will be required to maintain and monitor the dataset for FRBR compliance.

Initially the most useful entity for aggregation may be the work. This could be enhanced with selective harvesting and presentation of those expression (E) and manifestation (M) attributes that are the most useful to users and most likely to be consistently represented in the underlying NBD data. This is sometimes referred to as the upgraded or hybrid manifestation approach. Attributes that could be useful are listed:

• Form of expression - musical notation, musical sound, image
• Date of expression -
• Other distinguishing characteristics (E) - 'arranged' or 'version' statements (these are harder to extract but valuable for some categories of materials)
• Language of expression
• Edition (M) - second, revised, version 2.0
• Target audience (W)
• Date of publication/distribution (M)
• Terms of availability (M)
• Medium of performance (E)
• Type of score (E)
• Scale (E)
• Projection (E)

Future strategies

The Library is planning to implement an FRBR interface to the NBD as a part of the Kinetica Redevelopment Project. There are a number of issues to address in undertaking this work.

While FRBR is a conceptual framework, in developing the new interface Kinetica staff will explore the issues as a practical applicator, building upon the activities of LC, RLG and OCLC. Standards issues to be resolved include the adoption of data schemas in which to encode an FRBR database, the consideration of exchange mechanisms for FRBR data, the revision of international cataloguing standards to accommodate FRBR and the promotion of network standards to encourage best-practice data input to maximise the exposure FRBR entities within MARC records.

For the ANBD, with many million records, re-cataloguing is not an option. By utilising existing data consideration must be given to the extent to which an implementation that exploits all the FRBR entities: works, expressions, manifestation and items, or one that establishes a hybrid approach: work/expressions and manifestations, or works and manifestations will be delivered. This will be determined by the available MARC21 data. When FRBR becomes a concept embodied within cataloguing codes and bibliographic standards and library systems a different approach will be required.

Successful application of FRBR to the ANBD will involve extensive programming and some degree of human intervention to address issues in more complex record relationships, such as with Waltzing Matilda. Through the analysis and research reported in this paper the information required to take advantage of this new development in record storage and display has been established. The next steps will truly enable Kinetica users to more easily find resources in all forms and meet the information needs of end users and librarians.

Bibliography

Hickey TB, O'Neill ET and Toves J 2002 'Experiments with the IFLA Functional Requirements for Bibliographic Records (FRBR)'. D-Lib Magazine 8, 9 (September) http://www.dlib.org/dlib/september02/hickey/09hickey.html

Library of Congress 2003 FRBR Display Tool http://www.loc.gov/marc/marc-functional-analysis/tool.html

National Library of Australia 2003 Directions 2003-2005, Canberra http://www.nla.gov.au/library/directions.html

National Library of Australia 2003 Who'll come a Waltzing Matilda with me?, Canberra http://www.nla.gov.au/epubs/waltzingmatilda/index.html

Biographical information

Bemal Rajapatirana is currently senior analyst, Bibliographic Sub-systems, Kinetica's Development Project. More recently she has held the positions of manager of the National Bibliographic Database and the National Chinese Japanese Korean database. Bemal has also chaired the Kinetica Subject Headings Review Panel, and participated in a number of Kinetica's Expert Advisory Groups. Her interests encompass the exploration of emerging standards and in particular those that impact on data exchange and show potential for broadening the base of the National Bibliographic Database.

Roxanne Missingham is assistant director general, Resource Sharing Division, National Library of Australia. In this position she is responsible for the Kinetica service, Document Supply, PictureAustralia, Australian Library Gateway and CASL Consortium. She has a long career in libraries and IT focused on the development of digital delivery and digital services. She has been a library educator, library manager and researcher. She is a member the ALIA Board of Directors.