Facet Analysis: The Evolution of an Idea

Abstract Facets are widely encountered in information and knowledge organization, but there is much disparity in the use and understanding of concepts such as “facet,” “facet analysis,” and “faceted classification.” The paper traces the history of these ideas and how they have been employed in different contexts. What may be termed the classical school of faceted classification is given some prominence, through the ideas of Ranganathan and the Classification Research Group, but other interpretations are also explored. Attention is paid not only to the idea of what facet analysis is, and what purpose it serves, but also the language utilized to describe and explain it.


Introduction
The idea of facets as a basis for organizing information is now widespread and is generally regarded as a very powerful tool for improving knowledge representation and search. Facets pervade almost every area of information organization and management, and they have been applied to all kinds of indexing and retrieval systems, with all kinds of different purposes, from traditional classifications to ontologies, via thesauri, controlled vocabularies, commercial enterprise search, and discovery tools. 1 Hjørland has described facet analysis as "probably the most distinct approach to knowledge organization within Library and Information Science (LIS), and in many ways it has dominated what has be termed 'modern classification theory.'" 2 The understanding of what a facet is varies considerably from one context to another, and the language that is used to describe facets shows very little consistency or consensus on use. In different information sectors there may be hardly any correspondence between technical terminologies, with many different terms used to denote the same or similar phenomena. There is confusion as to the exact meaning of terms such as "facet," "faceted classification," and "facet analysis," and to a great extent they are used interchangeably with an apparent lack of concern for any technical differences. "Faceted classification," for example, may be used narrowly to describe a particular knowledge organization system constructed by means of facet analysis, or it may be applied more broadly to the whole activity of organizing information using facets. "Facet" itself appears only relatively late in the library and information science literature, although the general idea of an analytic approach to domain information, what we might consider to be facet analysis, is present long before mid-twentieth century writers' attempts to formalize the theory. This spontaneous emergence of facet analysis is very striking, even though it is not named as such, and seems to be indicative of a "faceted" mindset, long before the term "facet" comes into common usage.

Historical origins
Facet analysis is generally regarded by the Library and Information Science community as having its beginnings in India, in the work of S. R. Ranganathan and his followers, and, later, in the work of three distinct groups, in the United States (the Classification Research Study Group), the United Kingdom (the Classification Research Group), and India (the Indian Library Research Circle), who continued to work on and develop Ranganathan's theories. In particular, the United Kingdom Classification Research Group created a very distinctive and influential strand of facet theory, which is further discussed below.
It is difficult in this traditional form of facet analysis to separate the notion of facet from that of category, and, to a considerable degree they are used interchangeably. If we look for examples of the categorical analysis of concepts, there are a number to be found, often going back some hundreds of years. Attempts to create conceptual classifications using a categorical method very similar to facet analysis, and resulting in a faceted classification type structure, have been discerned in the writings of Hugh of St. Victor (1096-1141), 3 Ramon Lull (1232-1315), 4 Cyprian Kinner (d. 1649), 5 and Nicolas de Condorcet (1743-1794), 6 and particularly in the artificial languages of the seventeenth century, such that of George Dalgarno, 7 and, most comprehensively, in the work of John Wilkins (1614-1672). 8 Often the conceptual classifications created by these analyses are accompanied by ideas of combinatorial techniques to express complexity in a concept, 9 much as would be explored in the twentieth century. All of these classifications employ the language of categories, rather than facets, and owe something to the classical tradition of philosophic classification, principally Aristotle's Categories which are substance, quantity, qualification, relative, where, when, being-in-a-position, having, doing, being-affected. 10 The early twentieth century The idea of categories and categorical analysis first became applied to professional information work in the early twentieth century, 11 although the term "facet" was as yet not introduced. Part of a more general interest in analytical approaches to information organization and representation, this was largely occasioned by the increasing volume and subject complexity of documentation. Two major examples are the categorical tables used by James Duff Brown in his Subject Classification in 1906, and the categories employed in 1911 in the subject indexing system of Julius Otto Kaiser. 12 At around the same time basic analytico-synthetic, or "number" building, devices begin to be employed in library classifications more generally, notably in the work of Otlet and La Fontaine in the Universal Decimal Classification (UDC), and Henry Evelyn Bliss in the Bibliographic Classification, both through the use of auxiliary tables to represent commonly occurring concepts such as place, time, form, and language, and the facility for combining codes to describe complex content. 13 In neither of these two latter systems, however, is there any attempt at formally theorizing the structure of a discipline or domain; they are mainly concerned with more accurately representing compound subjects. Kaiser is much closer to the Ranganathanian school since his categories (concrete, process, and country, i.e., place, location) are used to analyze the vocabulary of a subject area on a broader basis and have a very similar feel to the fundamental categories that Ranganathan will later use. Svenonius proposes Kaiser as the true originator of facet analysis: "He was the first to recognize the usefulness of facets in the construction of expressions in a synthetic index language," 14 and it is hard to deny that his categories appear to foreshadow Ranganathan's. Brown's categorical tables, on the other hand, are less structured and more akin to the auxiliaries of UDC or Bliss, albeit more comprehensive in scope. Given the simultaneous emergence of these related ideas, Hudon 15 suggests that "Ranganathan did not invent the concept of facet; the idea was definitely 'in the air' at the time."

Ranganathan and the formal theory of facets
What we generally understand by the formal theory of faceted classification does indeed emanate from the work of Ranganathan in the 1930s and onward. It is likely that he was thinking about the idea of facet analysis as early as 1924 when he was a student in the library school at University College London, 16 was intrigued by the problems of classification, and came across the engineering toy Meccano, which is often said to have inspired him; 17 but at that stage his ideas were rudimentary and there was no vocabulary in which to express them. Beghtol 18 states that "the origin of the term 'facet' in the technical sense in bibliographic classification systems is unclear," and this is undoubtedly the case, but it is widely accepted that the term "facet" originates with Ranganathan although it seems not to occur in the literature before the 1950s. Ranganathan did not use it in his first work on classification, although the central mechanisms of facet analysis are present at this time. Initially he used phrases such as "string of characteristics. " Hudon 19 suggests that the first occurrence of the term facet was in 1944, "presumably in Library Classification: Fundamentals and Procedure." However, the need for clarity was obviously on Ranganathan's mind, as Satija 20 tells us that he advocated strongly for the adoption of technical vocabulary, and that he was responsible for "the introduction of new concepts like 'facet' , 'phase' , 'isolate' , 'focus' and many more." 21 What is essential to faceted classification is the use of fundamental categories as a tool for the analysis of concepts in a domain. Ranganathan's fundamental categories represent an advance on Kaiser's work, since they move from three basic categories (concrete, process, and country) to five (personality, matter, energy, space, and time). Ranganathan understood a category to be "a form or class of concepts, varying from subject to subject, into which isolates can be grouped," and a facet to be "the subclasses of a basic class corresponding to a single fundamental category." 22 This concurs with definitions supplied by Vickery, 23 who states that the "essence of facet analysis is the sorting of terms in a given field of knowledge into homogeneous, mutually exclusive facets, each derived from the parent universe by a single characteristic of division," and by the second edition of Bliss's Bibliographic Classification, BC2, where "[a] facet may be defined as the total set of subclasses produced when a class is divided by a single broad principle," 24 the single broad principle being a fundamental category, such as place or process. In this sense, the idea of a facet cannot be uncoupled from the idea of a category, and indeed they are often more loosely referred to as such, e.g., the property facet, or the operation facet.
A number of writers have identified a lack of rigor in Ranganathan's thinking. Foskett 25 in particular is critical of Ranganathan for his failure to provide an entirely satisfactory explanation of the categories, often resorting to ostensive definition, that is by the use of an example as illustration. This was particularly the case with the personality category, which he described as ineffable and unanalyzable. It is widely agreed that his writings are difficult to understand, and Raghavan records that his language has "been the subject of criticism by some, particularly western authors." 26 Spiteri 27 observes that "Ranganathan wrote a 'manual' for facet analysis in his Prolegomena to Library Classification in which he outlines in great detail the principles of facet analysis used in the design of classification systems. Although Prolegomena is readily available to LIS students, the same cannot be said for its contents," and "the semantic and syntactic structure of Ranganathan's language may serve to hinder easy comprehension of his principles of facet analysis." Palmer and Wells in their Fundamentals of Library Classification felt the need to simplify Ranganathan's ideas in order to bring them to a wider audience in the United Kingdom. In a review Shera 28 commends it as "an important book not only because it is a penetrating treatment of certain of the important problems of library classification but also because it draws attention, in clear and simple language, to Ranganathan's great contributions to the philosophy of bibliographic organization. The value of the work is materially increased because of the effort the authors have made to avoid an esoteric terminology (no simple task indeed)." The esoteric terminology is one part of a more general problem with Ranganathan's use of language. Satija 29 agrees that "difficulties of some of the readers, especially western, are not unreal." In suggesting some contributory factors, he includes the fact that English was not Ranganathan's mother tongue, that he was inventing and defining a new technical vocabulary for the subject, and that his cultural heritage was substantially different from western readers. The use of the Indian philosophic tradition and methods by Ranganathan is explored by a number of writers, including Gauri who states that "it is not difficult to find a parallel of his style in that of sutra (threads of argument) style popular in some of the early Hindu philosophical and religious writings. … All of the thirty-three canons of classification, twenty-three principles of helpful sequence and the guiding rules used in his total discourse are comparable to sutras." 30

Developments of Ranganathanian thinking: the Classification Research Group
The importance of Ranganathan's ideas is evidenced by the enthusiasm with which they were taken up by information professionals on an international basis and applied to the problems of information organization and retrieval in a wide variety of projects.
An examination of the literature on classification and information retrieval reveals the significance of the UK Classification Research Group (CRG). The emergence of Vickery as a major theorist must also be acknowledged. "Before the year 2000, the literature on facets and facet analysis was dominated by the works of Ranganathan himself, and of Brian C. Vickery," 31 which is supported by Tennis' assertion that "it is Vickery's 1960 and 1966 works, not Ranganathan's, that are often used as the introduction to (and often the end of the education in) faceted classification and facet analysis." 32 The CRG was set up by Vickery in the 1950s following the 1948 Royal Society Conference on Scientific Information. Speakers at the Conference had identified a serious problem with the dissemination of scientific information in the United Kingdom, particularly the results of research. A working party set up under the chairmanship of the physicist J. D. Bernal failed to make much progress, and Vickery was invited to form a group of fellow professionals to address the problem. In LIS terms, the rest is history. Their manifesto, published in the Library Association Record in 1955 was to make "faceted classification the basis of all information retrieval," 33 and it is perhaps at this stage that the terms facet and faceted classification take center stage. Despite their adherence to Ranganathan's principles, the CRG version of facets differs in some significant respects from his.
Tennis has made several studies of Ranganathan's developing methods, and compared these to Vickery's, where he detects a quite distinctive approach. Tennis writes that "Vickery's exposition of faceted classification and facet analysis were more parsimonious than Ranganathan's." 34 In the same vein, Spiteri's comparison of Ranganathan and the CRG principles 35 points out how complicated Ranganathan's ideas of rounds and levels are when compared to the CRG's standard citation order. Certainly, Vickery's approach was not positioned within the complexity of a largescale theory of classification, and more directly aimed at the problem in hand, so his method of designing a faceted classification was different. Much of the work of the CRG was directed at the construction of special schemes, and they afforded much more weight to the requirements and conditions of the local subject domain. It is within this context of special schemes that an explicit and detailed methodology for the construction of faceted schemes is documented, primarily in the works of Vickery. 36 A very significant difference from Ranganathan was the identification of a much larger set of fundamental categories (and hence facets), and the notion that additional categories might be generated by the analysis of any given field. Ranganathan's categories of personality, matter, energy, space, and time (collectively known as PMEST) were considered by the CRG to be limited, and they proposed a larger number. Vickery as early as 1957 expressed "a preference for terms such as Thing or Substance, Part, Property, Process, Action and Operation," 37 and in 1960 suggested a list of thirteen: substance (product); organ; constituent; structure; shape; property; object of action (patient, raw material); action; operation; process; agent; space; and time. These categories map to the smaller set of PMEST, but they are not all fundamental in the same sense as PMEST; the same concept may be a constituent, object of action, patient, or agent, according to the environment or the document in which it occurs. It is clear that the same phenomenon also was present in the Colon Classification (for example the use of place as personality in history) without being further explored at the theoretical level. 38 Mills, although at a much later date, makes the distinction between "true categories (Time, Space, Matter, etc.) and … relational categories (Kind, Part, Agent, etc.)." 39 However, Vickery, in personal correspondence with La Barre, stated that he "was not aware of this distinction and [did] not understand it." 40 Although a general consensus can be discerned among most of the members of the CRG, there were minor differences even within that consensus, and they did not in any case all speak with one voice. Austin 41 followed the line of "freely faceted classification" in which concepts are not bound to a discipline, and combination does not necessarily follow a prescribed citation order. A more radical departure was Farradane's relational indexing, 42 where, effectively, the relationships between terms (called relators or operators) were made explicit, and they were the basis of the classification rather than the concepts, or isolates, which were semantically neutral. Mills says of this "[a]ssigning terms to categories is a deductive approach to concept organization, and it may be noted that one member of the CRG advocated and developed an inductive approach." 43 However "[c]lassifications resulting from Farradane's system proved to be remarkably similar to those of faceted classification." 44 There is no mention at all of facets in Farradane's original work, although his 1952 paper discusses Ranganathan's use of facets at some length, with the conclusion that some features of the "Colon classification introduce [] a large degree of illogicality." 45 It has been mentioned above that the founding of the CRG was prompted by the need for efficient documentation in the realm of science, in particular the necessary development of a tool to support scientific research and dissemination at the time of the Cold War. In the 1960s the Group was awarded a grant by NATO to further this work (which ultimately resulted in the creation of the PRECIS indexing system), and this interest of governmental agencies was replicated in the United States by research at the George Washington University, funded by the U. S. Air Force Office of Scientific Research. A paper in the Classification Society Bulletin 46 demonstrates the use of facet analysis in this work and cites both Ranganathan and Vickery. Interestingly it discusses the notion of primary and secondary facets, as "developed by Vickery as a modification of a scheme developed by Ranganathan." 47 There seems to be no reference to these primary and secondary facets in the literature of CRG members, other than Mills' later reference to "true" and "relational" categories, so this may be an innovation on the part of the American researchers.
In summary, Beghtol writes that "[t]he influence of the classification work of the CRG, which continues to the present day both in the Minutes of their meetings and in papers explaining, among other things, the details of the facet approach …, cannot be overstated." 48

Developments of Ranganathanian thinking: facet analysis across the world
Despite the apparent dominance of the Ranganathanian school, at its beginnings, facet type approaches and analysis based on categories were in fact very widespread globally. La Barre 49 identifies several distinct schools of thought, even within what may be described as the classical tradition, with established research groups functioning in the US (Classification Research Study Group), and India (Indian Library Research Circle) as well as the United Kingdom Classification Research Group. Within these groups there were different understandings and interpretations of the ideas of categories and facets, and the process of facet analysis, some of which may be due to their "geographical and cultural separation. " 50 One might add to this set the more recent Italian research group, based within ISKO Italia and led by Gnoli, which is focused on freely faceted classification, 51 a development of the ideas originated by Austin in the CRG.
From the 1950s onward, there was a great deal of research taking place in mainland Europe, and the CRG had an American equivalent in the research group at Case Western Reserve, where the work was led by Allen Kent and J. W. Perry and resulted in their Semantic Code Dictionary. 52 The European work (and much in other parts of the world including the USSR, China, and Japan) is covered in some detail in de Grolier's work on the use of categories in classification and indexing (1962). 53 De Grolier lists research in eighteen different countries and by twenty-three separate individuals and organizations. France is particularly prominent with Pagès, 54 and Gardin 55 noted as important. While Ranganathan identified only five fundamental categories, and the CRG thirteen, the variety of categories or facets employed across these projects is much greater, and the facet analysis might be regarded more as a methodology or a process for conceptual analysis than a specific model for building a classificatory system. Kent and Perry's semantic code is particularly rich in categories, including (among others) material, component, product, agent or instrument, part, property, process, condition, location, and time, along with numerous role indicators and relationships such as influence and direction; these they refer to as "analytic relations," although the parallels with facets are very evident. Pagès' coded analysis includes a lexicon containing twenty-two categories of subject oriented content, exhibiting some of the general categories we have come to expect, including processes, properties, operations, phenomena, space, and time. Although most of the mid-twentieth century work is aimed at conceptual coding of subject content in the sciences and technology, with retrieval as the main objective, some research is humanistic and intended for primary rather than secondary sources. Gardin essayed a very close analysis of texts as diverse as Mesopotamian tablets, the Qur'an, and mythological narratives, and also considered aspects of cultural objects including ornamentation and iconography, in addition to the semantics of text, which would later come to the fore in the faceting of digital objects. Broughton observes that "Unlike previous exercises in analysis Gardin did not use any a priori group of categories or classificatory codes to manage the text, but extracted his concepts and categories on a strictly empirical basis. A subsequent analysis of myths had the aim of classifying the structure of the myth, and to permit a comparison of structural models, and finally to constitute a universal grammar of myths, and a catalogue of fundamental representations, or mythemes. These activities of modelling and classification building 'from the ground up' are so striking, that it is surprising to find so little reference to them in more recent work." 56

Facets in research: analysis and coding
In addition to its use as a method of dealing with complexity in subject content, facet analysis was also a vehicle for understanding the conceptual nature of subjects and subject domains, and another important dimension of classification research at this time was the establishment of a general theory of classification and indexing, as a foundational requirement of the newly emerging disciplines of library and information science. Most modern writers, such as Svenonius 57 credit Ranganathan with this broader achievement of creating a properly scientific basis, both for library science and for knowledge organization more narrowly, the latter of course largely dependent on his ideas about faceted classification; it is the case however that he constantly doubted the validity of what he was proposing.
In addition to forming a key component in library and information science theory, facet analysis also has some merit as a theory developing methodology in its own right. The part that facet analysis plays in creating a model of a domain is critical to theory formulation, and it has some commonality with other contemporaneous theory development tools such as grounded theory, which rely on the analysis and coding of text to reveal underlying conceptual structures. In their exposure of the concepts, structure, and relationships in a domain, "classifications are very much like theories. " 58 Foskett also references generic methods of text and data analysis in the social sciences in his discussion of the use of categories in revealing patterns and formalizing thought. 59 Facet analysis also appears to emerge independently of library science as a named approach in the realm of social sciences in the work of Gutmann, 60 which has been observed and referenced in the library science literature. 61 Gutmann defines facet in his own discipline thus: "By a facet we shall mean a set that is a component of a Cartesian product. No standard term for this appears in the mathematical or other literature. Experience shows the need for such a term, to distinguish the idea involved from related but often radically different concepts denoted by 'dimension,' 'factor,' 'element,' etc." 62 In a masterly analysis of the work of Guttman and Ranganathan, Beghtol concludes that "[i]n general, the discovery of facet concepts seems to have served like functions in both disciplines" and "in bibliographic classification theory, the problem was to make classification systems more conceptually flexible for describing the increasingly complex subjects of documents, and in behavioral research the problem was to increase the elasticity of quantitative methods to accommodate the description of qualitative data that arose from social science research. The facet concept proved exceptionally useful in both the disciplines in which it was discovered." 63 There is apparently no contemporaneous reference at all between documentation in the two disciplines, and it is difficult to trace their use of facets back to a common source, although Ranganathan's mathematical expertise may have meant he was familiar with such sources. Just as the notion of a facet was "in the air" in early twentieth century information science, so it may have been circulating in a broader disciplinary environment.
A little has been made of parallels between facet analysis and grounded theory in a single study by Star, 64 although her paper concentrates more on the similarities between the two contexts in which they were developed and makes less of common features in the methodology. Star does, however, stress the role of both as theory building methodologies, which is a key feature of grounded theory, but perhaps less commonly acknowledged for facet analysis. In terms of the epistemological foundation of facet analysis she somewhat surprisingly identifies it as an empirical method: "in both cases the categories are empirically discovered in an almost self-contradictory fashion" and "both grounded theory and faceted classification began as reform movements against powerfully entrenched a priori schemes. " 65 Facet analysis and grounded theory differ in their interpretation of categories and coding; Ranganathan's categories he regards as generalized and constant whereas grounded theory's emerge from the context and are context dependent, and they may be very numerous. Star tells us that Glaser (one of the originators of grounded theory) 66 "identifies eighteen families of theoretical codes including Process-stages, phases, transitions, ranks, etc.; Degree-limits, ranges, amounts, etc.; Dimensions-elements, pieces of, properties of, slices, segments, etc. (part-whole relations); and Ordering (including temporal ordering). 67 Kwasnik also writes about the role of classifications in general as theory forming tools. The example she provides of organizing a large amount of unstructured data is not dissimilar to grounded theoretical methods of analysis and coding. "The "inductive" discovery of the facets, through the filter of the researcher's analytical techniques, was an attempt to express the description of the classification of documents in the respondent's terms or categories rather than in terms dictated by a deductive model. This is reminiscent of the CRG's policy of allowing facets to emerge from the subject under examination 68 and is in marked contrast to Hjørland 69 who considers facet analysis as the a priori approach par example, grounded as it is in rationalism. For Kwasnik's purpose, the technique of facet analysis "was invaluable. … it can be seen that the fundamental categories are not so different, at least in underlying principle, from those outlined by Ranganathan." 70 If the links between facet analysis and grounded theory are rather tenuous, those with systems theory are more obvious, and indeed Foskett references Bertalanffy's General Systems Theory, 71 both in his work on classification in the social sciences, and in the 1970 publication Classification for a General Index Language reviewing recent research by the CRG. 72 The parallels are very close to soft systems methodology, as devised by Checkland 73 and developed by Smyth (1975), 74 and aimed at modeling systems domains, just as facet analysis can be employed to model subject domains. In soft systems methodology, the mnemonic CATWOE represents the analysis of a system using contextual elements which are analogous to categories. These are: customers, actors, transformation process, worldview, owner, and environment, and there are some clear correspondences here, not only with the method of categorical analysis, but with standard facet analysis categories such as entity, agent, patient, process, property, and so on.

Facets in information retrieval
In the mid-twentieth century, faceted classification was seen as a way to represent and to organize, in a predictable manner, the increasingly complex content in documents. The representation of different aspects of a subject through notation, and their coordination through the consistent application of an order of combination, or citation order, provided a reliable way to determine where complex topics would be located in the linear sequence. The use of facets, and the process of facet analysis, was primarily regarded as a means to the design and construction of conceptual systems for classifying and indexing, Although at the outset this was as an aid to retrieval in physical information collections, the role it might play in mechanized retrieval was appreciated early on, and several papers at the 1958 International Conference on Scientific Information stressed this potential, although eventually there would a split in the CRG between those more interested in the creation of classifications for organization, and those who focused on information retrieval more narrowly. Interestingly, this divergence also occurred in the Center for Documentation and Communication Research at Case Western Reserve University in the USA. 75 The rise of information technology in information work was important in enabling a move from the narrower confines of library science to the broader purpose of information retrieval, and the conjunction of the two resulted in new understandings of the idea of facets. Naun 76 emphasizes the very different requirements of tools for the physical collection and for its online equivalent, and the more significant role that facet analysis might play in the latter: It was not Ranganathan's work, in isolation, which led to the upsurge of interest in classification theory. The post-war period also saw the rapid rise of computer technology, with the attendant recognition of its potential in information retrieval. Facet theory attracted interest because it offered a principled and systematic way of organising knowledge, a way which would be amenable to computer-based techniques. … It was pointed out that the analytico-synthetic method permitted the handling not only of brute subject headings, but of thesaurally relatable terms; and not only of Boolean aggregates of terms, but of syntactically significant concatenations of terms. As long as classification served only to determine shelf arrangement, and subject-matter was accessible only through laboriously constructed card entries, these remained somewhat academic distinctions. But with the new power of computers to manipulate data these ideas took on much greater significance.
Just as the idea of facets had been "in the air" before the time of Ranganathan, so it took additional momentum from the new online environment, albeit this time with the advantage of a sound philosophical and theoretical basis. In general, the foundation years of faceted work placed the emphasis on a conceptual approach, analyzing and organizing concepts into categories or facets and thence into systematic structures. In Europe and Asia the application of facet analysis still tends toward the classificatory, with the emphasis on organization and representation in ongoing work on the Colon Classification and the second edition of the Bibliographic Classification, both continuing the original tradition of Ranganathan and the CRG. The Universal Decimal Classification also adopts this approach in a program to rationalize its structure, and the recently introduced Integrative Levels Classification 77 shows another more complex and nuanced version of a faceted system that includes relationships among its facets.
It is important to note that, in the classical school of facet analysis, the term "facet" applies only to aspects or dimensions of a subject or subject field. Categories, or facets, such as place, time, or agent, refer only to those concepts as an element of the subject, e.g., "library science in India," "medieval cosmology," or "heart disease in women," and not documents produced in a particular period or place or by a particular person or organization. This is a marked difference from what comes later in the development of the faceted approach, particularly in its digital applications.

Faceted thesauri and vocabularies
From the 1980s onwards much of the underlying theory of faceted classification was recognized to be relevant to vocabularies as well. The controlled vocabulary per se includes one aspect (the interterm relationships) which is apparently absent from the faceted classification. Vickery states that "to combine two facets is to imply a relation between them, but it is the explicit naming of relations between terms that is another theme in knowledge organization, found in thesauri and ontologies but not in facets as such." 78 Nevertheless, it was recognized that these relationships implicit in the classification, and usually very evident in the schedule display, could become explicit in its alphabetical equivalent, and the idea of the faceted thesaurus was conceived. One of the earliest exercises in creating a faceted thesaurus is Thesaurofacet, 79 and the method is now well established and detailed in various sources for thesaurus construction published since then. 80 Almost all the information necessary to the creation of a faceted thesaurus to ISO standards is inherent in a faceted classification of concepts in the same domain. In the creation of a faceted thesaurus using a faceted classification as a starting point, the systematic structure is used to determine explicitly the broader BT and narrower NT terms (subordinate and superordinate classes within facets in the classification), and other terms which do not conform to these criteria, the associative or related terms, RTs, which usually exhibit relationships between terms in different facets. The display of synonyms in class headings in the classification also help with the identification of preferred and non-preferred terms and the USE and UF relations. This interdependence of the faceted classification and its accompanying thesaurus is so logically sound that it is possible to generate the thesaurus automatically without human intervention. 81 The phenomenon is a further argument for the strength of facet analysis as a general knowledge organization theory.
In the documentation for the most recent British Standard 82 for the construction of vocabularies, which itself underpins the current International Standard, it is proposed that the faceted thesaurus reveals a variety of more nuanced associative relationships, such as those of an operation and the agent performing it, or a process and the product which results. To date, however, these additional relationships have not been incorporated into any published thesaurus. Although a faceted thesaurus is often generated, either manually or automatically, from a faceted classification, concerns about vocabulary control and authority come into play here, which had not been an issue for systematic tools, and most faceted thesauri of the conventional type will require manual editing.
It has been the convention for the faceted thesaurus to contain both the traditional alphabetically arranged format, and the systematic structure from which it is derived. When presented online, it may be difficult to differentiate a faceted thesaurus from a faceted classification, particularly where the thesaurus supports a browse function, and the two dimensions of the faceted thesaurus are both presented as alternative search tools. Perhaps the best known of the online thesauri, the Art and Architecture Thesaurus (AAT), looks much more like a faceted classification if viewed through the facet hierarchies, and it has a notably rigorous structure in its display of arrays within the facets. Interestingly, the AAT site once contained a lengthy discussion of Ranganathan's theories and the way in which AAT was based upon them, but in recent times this feature has been abandoned.
The thesaurus is conventionally the tool used by information professionals for indexing content, but it is increasingly offered to the end user as means of retrieval, that is a search thesaurus. When the thesaurus is employed as a search thesaurus it is often the systematic display, or classification, that supports the search function. Early work on visualization in search tools by Pollitt 83 showed how drop-down menus revealing individual facet hierarchies could be used to frame complex searches, and the same method is now widely used in all kinds of subject related search tools. Tudhope and his team 84 have also created various thesauri for cultural heritage collections which use the same technique, but in that case with a degree of automatic query formulation and modification based on the faceted structure. Tudhope and Binding state that "today's standards make clear that the thesaurus is concept based and that the systematic display structured by semantic relationships is crucial, together with access mechanisms based on the semantic structure." 85

Facets on the Web
It should be acknowledged that the understanding and use of the term "facet" in the faceted thesaurus is not very substantially different from that employed by Ranganathan and his followers. The same is not necessarily true of a new wave of online tools described as faceted which were devised by web designers and information architects in the late 1990s and onwards. With respect to the web-based thesaurus, Tudhope allows that while some prominent thesauri are faceted, this is not always the case: "[a]lthough most thesauri contain different hierarchies, they would not be considered faceted. In such cases the hierarchies are not based on a rigorous process of facet analysis." 86 And they "can include facets that are essentially pick lists and there is usually little notion of the semantics of combining facets. However, this simple facet treatment can yield attractive browsing interfaces for websites." 87 This phenomenon has been comprehensively recorded and analyzed, particularly by US authors, in a number of key publications, notably Adkisson 88 and La Barre. 89 Alongside the scholarly commentary on faceted online search there also arose a parallel "school" of faceted classification with its own pundits, resources, and key texts. Most of these practitioners were a very long way from the classical school of faceted classification, and usually lacked any formal education in library science. This is particularly true of the North American tradition, which, while initially rooted in the Ranganathanian school, diverged over time. As La Barre says "the principles and techniques seem to have steadily rumbled along over time, never entirely forgotten and often hidden in systems that made only passing recognition. " 90 La Barre notes that despite the early US work on faceted classification, in the 1970s and 1980s there were fewer voices in the debate, and that "one result of the paucity of explicit discussions of facet theory in North America during the 1970s and 1980s, is that contemporary North American understanding and applications of facet analysis and faceted classification are varied and do not always reflect awareness of heritage principles." 91 Wild, Giess, and McMahon also raise the matter of intractability and lack of access to the canonical literature when facets are considered by those outside the LIS sector: "The divergence in faceted approaches has several roots. One is the difficulty of obtaining the original work. Once obtained there is the difficulty of interpreting Ranganathan's dense style. Finally, there is a more general tendency to apply notions faceted in different contexts with different philosophical assumptions and pragmatic aims." 92 La Barre also emphasizes the distinction between a classification of knowledge, and a classification of objects. This distinction had been made much earlier when the UK CRG discussed the "Chinese plate syndrome, " 93 and it is central to the division between faceted classification for bibliographic purposes and web faceted classification. Interestingly, the CRG concluded that it was possible to establish what the Chinese plates were, but not necessarily what they were about. Vickery adds another dimension to the argument when he says: "on the Web there is no direct classification of 'objects' (as there could be in a compartmented box to keep in orderly arrangement different shapes and sizes of nails and screws; or in a natural history museum). Both objects and subjects on the Web are represented by descriptions in symbolic terms (words, phrases, codes, icons, etc), which point to objects or documents. " 94 A faceted system in this latter (object) sense is often only equivalent to a taxonomy, intended to organize some particular entity or product according to its properties or attributes. The process of analysis using a larger set of categories is entirely absent here, and such systems lack the depth that is implicit in the notion of "subdivision of a subject field" which must contain not only entities, but also materials and parts, processes and operations, agents and products, and also space and time. Such examples are very significant in the development of facet practice because they introduce an idea of the facet which is very different from the traditional one. Whereas traditional facets map to categories and are essentially conceptual and abstract in nature, the web faceted classification maps to properties that are practical and non-conceptual in nature. La Barre gives the nice instance of the "facets" of a detergent which are "'brand name, ' 'form, ' 'scent, ' 'agent, ' 'effect on agent, ' and 'special property. '" 95 Merholz, an early leading exponent of web faceting posits the example of "Toys, for example, [which] lend themselves to a faceted classification, with the facets being things like, 'Suitable Age, ' 'Price, ' 'Subject Type, ' 'Brand, ' and even 'Character' (like Barbie or Elmo). " 96 Possibly the earliest website designed in this way, and also the most influential in the field, was Wine.com 97 which allowed a customer to select a label based on a variety of attributes.
Web faceted classification, as might be expected, uses a different terminology to discuss facets. Faceted search in a commercial context is often described as enterprise search, or parametric search. Browsing, as opposed to search, may be termed faceted navigation. The term attributes is frequently used to explain facets, and the subclasses of attributes are values. One website states that there are four kinds of facets: enumerative, Boolean, hierarchical, and spectrum. 98 One is reminded in part of the differences between the language of thesaurus construction and software engineering where quite different terms are used for identical concepts (broader/narrower relationship and parent/child relationship), but the disparities here seem on a different scale.
These applications essentially deal with aspects of the product, or the consumer, and not with dimensions of a subject; they have to do with objects and not with knowledge. They can be contrasted with more sophisticated systems designed for the indexing and retrieval of objects as opposed to documents; obvious examples are Gardin's work on archaeological artefacts 99 and Tudhope's on museum collections. But because the purpose is not to represent complex semantic content, relationships between concepts in the web faceted classification are not part of the description, as is the case with the faceted classification or faceted thesaurus.

Subject cataloging tools and headings
A very similar development of the idea of facet can be seen in the application of facet analysis to subject cataloging tools, where the objects in question are bibliographic records. The most prominent example of such a tool is the FAST (Faceted Application of Subject Terminology) project for the Library of Congress Subject Headings (LCSH). Those schooled in the classical tradition of faceted classification and who consider "enumerative" and "faceted" to be mutually exclusive concepts, will no doubt be surprised to learn that FAST "is an enumerative, faceted subject heading schema." 100 It has been pointed out that the two need not be in conflict, and that a scheme might be faceted in structure, but enumerative in display. 101 It is clear that the idea of 'faceted' is applied here in a very basic manner: "Recently, the ALCTS/SAC/Subcommittee on Metadata and Subject Analysis … recommended that LCSH strings be broken up [faceted] into topic, place, period, language, etc., particularly in situations where non-catalogers are assigning the headings." 102 "The first phase of the FAST development includes the development of facets based on the vocabulary found in LCSH topical and geographic headings and is limited to six facets: topical, geographic, form, period, with the most recent work focused on faceting personal and corporate names." 103 It is hard to see how, by any stretch of the imagination, one could facet a personal name in traditional facet analysis, and it is clear that some very different understanding is in operation here. It seems probable that the faceting of names refers really to the removal of names from more complex headings, and their 're-location; to a separate names facet (essentially equivalent to a name authority file), and that the FAST usage of faceting in essence means this deconstruction of pre-coordinate headings. There is little written about any underlying theory, and it appears that the exercise is largely pragmatic. One interesting contrast with the classical tradition is the continual emphasis on the semantic content and richness of LCSH, and the need to retain this in FAST, whereas commentators of the old school had expressed the belief that faceted classification concentrated too much on the syntactics of facet analysis.
The aim of the project was to create a simplified subset of LCSH that would have a much less pre-coordinated structure, and would be easier, especially for paraprofessionals or unqualified catalogers, to apply. It should be acknowledged that this was not expected to result in better resource description, retrieval, and user satisfaction, but because largely economic drivers (limited cataloging budgets, the scale of the cataloging workload, and lack of professionally trained staff) rendered the continued use of LCSH unrealistic. A UK study by Cuna and Angeli 104 acknowledges this tradeoff, and the consequences: It is important to recall that a shallow application of [facet analysis] and classification are not unique to FAST …, but common practice in the design of most next generation catalog interfaces. Many view this practice as merely a pragmatic and cost-effective way of dealing with the problem of leveraging existing metadata to populate the library catalog interface with facets. But facets derived in this way lack the semantic richness and formalisms necessary to organize topics into a cohesive and logical multidimensional structure that can truly support the user's exploratory search tasks.
A particular object was to reduce the complex syntax of LCSH, with the implication that the relationships between terms were less significant in FAST than in LCSH. The facets in FAST now consist of "nine distinct categories or facets: Personal names, Corporate names, Meeting names, Geographic names, Events, Titles, Time periods, Topics, and Form/Genre. " 105 Rather than coordinating terms from different categories, the terms are assigned post-coordinately, as separate headings, simplifying the syntax, but also losing some of the semantics implicit in a structured heading. Despite the stress on semantic richness, these facets are more closely aligned with the fields in a bibliographic record than with the categories in the semantic analysis of a complex content in a subject domain.

Faceted search and discovery
Nowadays, the library user may most commonly encounter facets as a dimension of discovery tools. In the UK these software applications are employed in almost all academic libraries and provide a structured support to searching the collection. The discovery tool shows features compatible with the FAST headings since it sits on top of the catalog software and relies on the content of the catalog to populate its structure, content which it draws from the separate fields in the catalog database. This is most evident when the discovery tool uses an interface that 'visualizes' the bibliographic record as a set of separate hierarchies, or facets, corresponding to the fields in the catalog database. Unlike the faceted search used on most commercial sites, the user must normally start with a simple keyword search box but is then able to modify the search results which are presented alongside the separate facets, by selecting additional concepts from their hierarchies. As might be expected, these facets consist of fields such as document format (ebook, article, etc.), location, currency (new books), date (of publication), author, language, and topic. Topic (i.e., subject) is only a small element of faceted search, and there is nothing of the complexity of subject faceting of the traditional kind. In the first manifestations of discovery tools, the concepts listed under topics were very evidently LCSH, evidenced by their pre-coordinate structure. An examination of my own institution's catalog today suggests that these have been replaced by FAST headings.
For most users it avoids the need to drill down into the catalog and offers an easier way to understand the bibliographic information than the older advanced search option. Critics, however, have seen the shortcomings in faceted search of this kind, both in its lack of theoretical robustness, and its usefulness. Hall 106 writes: General subject/topic facets are crudely implemented. A mix of controlled vocabulary (e.g., library of congress subject headings) and keyword terms, subject facets in OPACs combine people, places, actions, locations, time, entities and many more characteristics and categories that traditional facet theory teases apart into individual and fundamental categories. The facets which are most frequently observed across library catalogs may offer a list of OPAC-oriented fundamental categories, and a rudimentary division of facets into administrative, bibliographic and subject categories could provide a useful starting point for more detailed appraisal. However, many of these facets are equally likely to simply indicate a list of easily extracted metadata. As well as adequately describing the domain, facets must be relevant to the purpose of the information system and useful to the user. Facets like 'place of publication,' 'uniform title,' and 'local terms' (all of which do, occasionally, occur) are not useful for the majority of searches and in the case of Uniform Title and Local Terms are most likely meaningless to a non-librarian.

Ontologies and beyond
An ontology consists essentially of a taxonomy of concepts and their relations, and a set of inference rules that support the reasoning function of the ontology. While in traditional LIS, the facet would sit at the top level of the taxonomy, this is not the case in ontology. Gnoli et al., 107 citing Noy and McGuinness, 108 note that 'facet' is used in quite another sense, where "slots (sometimes called roles or properties)" are more like LIS facets, and facets (sometimes called role restrictions) are of another nature; "restrictions on slots can have different facets describing the value type, allowed values, the number of the values (cardinality), and other features of the values the slot can take." Here, the facet has a much more subordinate role in the taxonomic structure, more akin to the array, or characteristic of division in classification tradition.
The heirs of Ranganathan in Bangalore have applied his principles to ontology construction, where facets are central to the process of building the taxonomy. Just as a thesaurus was built using a faceted classification as a starting point, so in recent years ontologies have been engineered on the same basis. Ranganathan may be specifically cited as the inspiration for this, and his principles employed. 109 Dutta et al. use the terminology of Ranganathan (facet, characteristic, array), but in ways that do not necessarily conform to the original. For example: "a facet is a hierarchy of homogeneous terms describing an aspect of the domain, where each term in the hierarchy denotes a different concept." "Concepts inside a facet are arranged by characteristics, i.e., according to their distinctive properties. For instance, since both river and brook are flowing bodies of water (their characteristic) they are arranged in the same facet, i.e., body of water, and at the same level of the facet hierarchy. When arranged together, siblings sharing the same characteristics form what in jargon is called an array of homogeneous terms. " 110 The authors acknowledge that there are differences between their approach and the LIS method, and attribute this to the difference between the representation of subjects in documents, and the ontology of real-world objects where each concept may denote a class, an entity, a relation or an attribute.
More strikingly the phrase fundamental categories is introduced (probably uniquely in this area); the fundamental categories being: C classes of realworld objects; E instances of the classes in C; R the relations between entities and classes; and A qualitative/quantitative and descriptive attributes of the entities (there is a further distinction between attribute names and attribute values). Ranganathan's PMEST has become CERA. One can see the correspondences in the terminology, and in the thinking, but in the ontology the facet has become uncoupled from the category, and the resulting structure sits somewhere between the faceted classification proper and the taxonomy of faceted websites. The representation of relations, which may seem unique to the ontology, is not in itself new. In 1993, in the revision of the Mathematics class, BC2 introduced a new category of relations, although these are mathematical concepts in their own right (mappings, functions, equations), not inter-facet relationships of the entity-process or operation-agent kind. 111 Closer to the thinking of ontology engineers, is that of Farradane's relational indexing in which the concepts (called isolates) had no syntactic value, and the classification was of the relations between them.

Conclusion
Some seventy years after the CRG manifesto, and fifty years after Ranganathan's death, the influence of facets can be seen in every type of subject cataloging and retrieval tool: classifications, thesauri, subject headings, directories, ontologies, both in libraries and on the World Wide Web. Not only the design principles but also the theory can be seen to underlie the construction of more contemporary tools, and to inform information architecture on the web. Alongside this development come techniques for expressing knowledge organization systems in machine readable languages, where the rigor and logic of the faceted system proves to be a considerable advantage. Within these systems one can discern the different objectives of organization and navigation, and, in recent times, there has been a major shift from the former to the latter, perhaps mirroring the division between physical organization and retrieval which occurred in the 1970s. Today the faceted approach to knowledge organization is usually regarded as a way to rationalize and simplify the description of resources, and a means of providing a comprehensible structure in an information store, especially a database. Particularly important is the capacity to visualize the domain to aid user understanding. In this context, the idea of a facet may be considered more simply as any aspect of the data attached to bibliographic or other information objects, where subject may be only one among a range of attributes including author/creator, format, language, date of origin, and so on.
It is hard in a very interconnected world to trace the influences. Sources for the FAST project make little or no reference to Ranganathan and his school of thought; the Getty vocabularies did originally tell the story but have since removed it. There is little interchange between the library community and the web community; web designers making faceted classifications may know about Ranganathan but rarely cite any more recent LIS literature. While a direct legacy is seldom acknowledged the fundamentals of facet analysis seem inherent in most more modern approaches, and the idea of rigorous conceptual analysis underpins them. Language is undoubtedly a barrier, but it is unlikely that there will ever be consensus among these different factions. The major difference lies in the depth of the analysis and the number of categories employed, the subject faceting being rich and complex, the web faceting shallower, using few of the traditional categories. What might qualify as a facet is also very disparate, from the conceptual philosophical categories of Aristotle which strongly influence the Ranganathanian school, to simple properties or attributes of an object or product, to the fields in a database or the relations in an ontology.
It can be seen that facets have over time been applied to a range of tasks: physical arrangement, organization, retrieval, navigation, and visualization. They have also been instrumental in domain analysis, semantic analysis of texts and images, and theory building in library and information science. They probably constitute the most powerful intellectual tool in modern information management, and exist, either implicitly or explicitly, in almost all contemporary systems. However, the precise definition of a facet is always elusive; although the term was almost certainly coined by Ranganathan, he did not explain it clearly, and even his immediate followers varied in their interpretation. It should also be acknowledged that Ranganathan's thoughts about facets themselves evolved significantly throughout his lifetime, 112 from an initially undefined concept, via the formal canons and laws of library science, to the later idea that a facet is an element or component of a subject, no more specifically delineated than that. It is perhaps a useful label for something that is always "in the air, " a structured and analytical way of dealing with information that allows for more precision in search, and more accuracy and efficiency in retrieval.