Basic-level categories: A review

Article

Basic-level categories: A review

Journal of Information Science 39(5) 676–687 Ó The Author(s) 2013 Reprints and permissions: sagepub.co.uk/journalsPermissions.nav DOI: 10.1177/0165551513481443 jis.sagepub.com

Lala Hajibayova School of Library and Information Science, Indiana University Bloomington, IN, USA

Abstract This paper analyses selected literature on basic-level categories, explores related theories and discusses theoretical explanations of the phenomenon of basic-level categories. A substantial body of research has proposed that basic-level categories are the first categories formed during perception of the environment, the first learned by children and those most used in language. Experimental studies suggest that high-level (or superordinate) categories lack informativeness because they are represented by only a few attributes and lowlevel (or subordinate) categories lack cognitive economy because they are represented by too many attributes. Studies in library and information science have demonstrated the prevalence of basic-level categories in knowledge organization and representation systems such as thesauri and in image indexing and retrieval; and it has been suggested that the universality of basic-level categories could be used for building crosswalks between classificatory systems and user-centred indexing. However, while there is evidence of the pervasiveness of basic-level categories, they may actually be unstable across individuals, domains or cultures and thus unable to support broad generalizations. This paper discusses application of Heidegger’s notion of handiness as a framework for understanding the relational nature of basic-level categories.

Keywords basic-level categories; basic-level terms; categorization; handiness; Heidegger; knowledge organization and representation; theories of categorization

1. Introduction In his seminal essay entitled ‘How shall a thing be called?’ [1], Roger Brown observed that there is a basic level of vocabulary that parents use when teaching their children to name and categorize things. Why is it that the spoon is rarely referred to as anything but spoon and not as silverware or artefact; and why is a dime called dime and not metal object, 1952 dime or 10 cents? Conversely, this preferred level of naming systematically violates frequency of occurrence: a pomegranate is pomegranate instead of fruit. Brown claimed that such examples indicate that the preferred level of names is the level that is most useful in most contexts. Drawing inspiration from Brown’s observations and from Wittgenstein’s [2] discussions of language use, psychologist Eleanor Rosch [3, 4] developed the theory of basic-level categories. Based upon her empirical observations, Rosch suggested that a taxonomic system contains a basic level at which categories are most cognitively efficient [3]. In Lakoff’s [5] terms, these basic-level categories are ‘human-sized’ in that they depend upon how people interact with the objects represented by a category name: the way people perceive these objects and organize information about them. The phenomenon of basic-level categories has been studied across different domains. Psychology, linguistics and anthropology have developed various measures of perception, communication and behaviour that converge on the basic level of categories. The pervasiveness of basic-level categories has also been observed in library and information science studies: the prevalence of basic-level terms across different systems of knowledge organization and representation has been demonstrated by Green [6] and the use of basic-level categories has been identified in image indexing and retrieval [7, 8]. Even though the universality of basic-level categories in systems of knowledge organization and representation would seem to indicate their universality in indexing as well, basic-level categories may not be appropriate for representing specificity or for supporting collocation precisely because they have a high probability of occurrence.

Corresponding author: Lala Hajibayova, School of Library and Information Science, Indiana University in Bloomington, Indiana, USA. Email: [email protected]

Hajibayova

677

This paper reviews selected literature on basic-level categories, explores related theories, and discusses theoretical explanations of the phenomenon of basic-level categories.

2. The structure of categories Most of the distinctions we make in our lives are related to our mental ability to partition space and create categories by separating ‘things’ from the context in which they are embedded [9]. Categories are formations of the culture that are coded by its language at a particular point in time [10]. Rosch suggests that the main principles of category formation are cognitive economy and perceived word structure. She contends that cognitive economy is central to the primary functionality of a category system: ‘[T]he task of category systems is to provide maximum information with the least cognitive effort’ [10, p. 28]. Perceived world structure, in contrast, deals with the nature of information: the ‘perceived world comes as structured information rather than as arbitrary or unpredictable attributes’ [10, p. 28]. But how are categories formed? And why are some categories well-defined while others are fuzzy? Although these questions have been studied for decades, there are still no definite answers. The classical theory of categorization has been the dominant theory of category representation for decades, if not centuries. The classical theory is based on the notion that a category is determined by a set of defining criteria. According to Smith and Medin [11], there are three assumptions supporting the classical view: 1. 2. 3.

summary representation.; necessary and sufficient features; nesting of features or inheritance of defining features.

The classical view has been challenged by experimental findings regarding typicality effects, determinants of typicality, application of nonnecessary features and nested categories. One of the earliest studies that demonstrated the problems of defining features was conducted by Clark Hull in the 1920s [12]. As described in Murphy [13], Hull used novel visual forms to ‘predict’ what features would occur in all instances of a category because, for many categories of objects, the defining features are functional. The findings of Rosch [3, 14] further undermined the assumption that all category members are equally representative. Rosch’s early studies on colour [15, 16] indicated that there were focal colours that were most exemplary, linguistically codable and easily remembered; and she observed young children’s preferences for focal colours over nonfocal colours in a study of colour name development [15]. Focal colours were most shown, most matched and most chosen. Similarly, Rosch’s study of the colour categorization patterns of the Dani language speakers of New Guinea supported her earlier findings [14]. Even though Dani colour terms were restricted to two basic colours – mili (dark colours) and mola (light colours) – the study demonstrated that Dani subjects were able to ‘learn names for the presumed natural prototypes faster than for other stimuli even though the natural prototypes were not the central members of categories’ and that they chose the ‘natural prototype as the most typical member of the category even when it was actually peripheral’ [14, p. 348]. In an extended study of categories ranging from colours to physical objects, Rosch and Mervis [3] suggested that the internal structure of categories was comprised of prototypes or ‘clearest cases, best examples of the category’ [3, p. 574]. These findings led to the introduction of a new view of categorization known as prototype theory [17, 18]. Smith [18] has observed that, according to the prototype view, an ‘object is categorized as an instance of a . [category] if it is sufficiently similar to the prototype, similarity being determined in part by the number of [features] that the object and prototype share’ [18, p. 504]. Rosch’s notion of category prototypes was closely aligned to Wittgenstein’s notion of family resemblance. Indeed, Rosch and Mervis [3] argued that prototypes were an ‘empirical confirmation of Wittgenstein’s (1953) argument that formal criteria are neither a logical nor psychological necessity; the categorical relationship in categories which do not appear to posses criterial attributes . can be understood in terms of the principle of family resemblance’ [3, p. 603]. Thus, a category was not governed by definitions but rather by a ‘complicated network of similarities overlapping and criss-crossing: sometimes overall similarities, sometimes similarities of detail’ [2, p. 32]. For Rosch, as for Wittgenstein, family resemblance was a ‘thread twist[ed] fibre on fibre’ and the ‘strength of the thread does not reside in the fact that some one fibre runs through the whole thread – namely the continuous overlapping of those fibres’ [2, p. 32]. The notion of family resemblance also underlies a process of formation of prototype as part of the general process by which categories are formed [3, 19]. In a series of experimental studies, Rosch and her associates observed [10] that, in the perceived world, information-rich bundles of perceptual and functional attributes form natural discontinuities and that basic cuts in categorization were made at these discontinuities. Working from these assumptions, Rosch et al. [4, Journal of Information Science, 39 (5) 2013, pp. 676–687 Ó The Author(s), DOI: 10.1177/0165551513481443

Hajibayova

678

10] identified basic-level categories as most inclusive because they ‘carry the most information, possess the highest cue validity, and are, thus, the most differentiated from one another’ [4, p. 383]. According to Rosch and Mervis [3, p. 586]: Basic objects (for example, chair, car) are the most inclusive level of abstraction at which categories can mirror the correlational structure . of the environment and the most inclusive level at which there can be many attributes common to all or most members of the categories.

Basic-level categories provide a balance between informativeness and cognitive economy. Experimental studies suggest that superordinate categories lack informativeness because they are represented by only a few attributes and subordinate categories lack cognitive economy because they are represented by too many attributes [3]. A substantial body of research has proposed that basic-level categories are the first categories formed during perception of the environment, the first learned by children and those most used in language. In the years since Rosch introduced the prototype theory of categorization and the notion of basic-level categories, her experimental approach has been applied to a wide variety of domains: word use and free naming [20–22]; American Sign Language for the deaf [23]; the categorization of images [24, 25]; and environmental scenes [26, 27].

3. Basic-level categories 3.1. Studies of the basic-level categories Roger Brown is often identified as the first researcher to have theorized the existence of a basic level of categorization [5, 13]. In his seminal article, Brown [1] commented on how parents teach their children to name and categorize objects and argued that the ‘name of a thing, the one that tells what it ‘‘really’’ is, is the name that constitutes the referent as it needs to be constituted for most purposes’ [1, p. 17]: The dime in my pocket is not only a dime. It is also money, a metal object, a thing, and, moving to subordinates, it is a 1952 dime, in fact a particular 1952 dime with a unique pattern of scratches, discolorations, and smooth places. When such an object is named for a very young child how is it called? It may be named money or dime but probably not metal object, thing, 1952 dime, or particular 1952 dime. The dog out on the lawn is not only a dog but is also a boxer, a quadruped, an animate being; it is the landlord’s dog, named Prince. How will it be identified for a child? Sometimes it will be called a dog, sometimes Prince, less often a boxer, and almost never a quadruped, or animate being.

Brown contends that the names of things as they really are serve as true representations of the referent while other names given to the object are ‘recategorizations’ that represent ‘acts of imagination’ [1, p. 17]. It is these names of things as they are that presaged Rosch’s theories of prototypes and basic-level categories. In a series of experimental studies, Rosch et al. [4] observed that there was a particular level in a category hierarchy where the category possessed significant numbers of attributes in common and shared few characteristics with members of categories at the same level, had similar motor programmes governing interaction with individuals and could be generally identified by an average shape. Furthermore, Rosch’s study of the structure of categories showed that these basiclevel categories were: [the] most inclusive categories for which a concrete image of the category as a whole can be formed, to be the first categorizations made during perception of the environment, to be the earliest categories sorted and earliest named by children, and to be the categories most codable, most coded, and most necessary in language. [4, p. 382]

The pervasiveness of basic-level terms had also been observed in early studies of colour and plant categorizations. Berlin and Kay’s [28] cross-cultural study of 20 languages suggested that there is a set of basic colour terms (BLACK, WHITE, GREY, RED, YELLOW, GREEN, BLUE, PINK, ORANGE, BROWN and PURPLE) whose extension on the colour spectrum is similar across different languages. Berlin et al.’s [29] studies of how Tzeltal speakers living in Tenejapa, Mexico, classified plants suggested that they tended to name plants at the level of genus (e.g. oak) rather than species (e.g. live oak). Berlin [30, p. 17] argued that, ‘psychologically, generic taxa are highly salient and are the first terms encountered in ethnobiological inquiry, presumably because they refer to the most commonly used, every-day categories of folk biological knowledge’. He theorized that the ‘generic rank [provides] the core of the entire folk biological taxonomy’ and that it is at this level that ‘plants and animals appear perceptually most distinct to the human classifier’ [30, p. 24]. Journal of Information Science, 39 (5) 2013, pp. 676–687 Ó The Author(s), DOI: 10.1177/0165551513481443

Hajibayova

679

Based on studies of everyday language, Lakoff and Johnson [31, 32] suggested that conceptual structure is meaningful because it is embodied, that is, it arises from, and is tied to, our perceptual bodily experiences. For Lakoff embodiment is coupled with ‘collective biological capacities and . physical and social experiences [of individuals] as beings functioning in . [their] environment’ [5, p. 267]. Lakoff and Johnson assume the existence of an external structure in ‘our bodily experience’ [5, p. 267], which arises from the structured ‘preconceptual experiences’ [5, p. 267] of individuals. Lakoff [5] suggests that there are at least two kinds of structure that are implicated in preconceptual experiences: basic-level structure and kinesthetic image-schematic structure. Basic-level structures comprise categories that represent the convergence of gestalt perception, the capacity for bodily movement, and the development of rich mental images, while kinesthetic image-schematic structures are the external structures that constantly appear in everyday bodily experiences – structures such as paths, containers or front–back and part–whole relationships [5, p. 267]. Lakoff and Johnson also theorize that, through the use of metaphors and the projection of basic-level categories to superordinate and subordinate categories, individuals extend these basic-level and image-schematic structures to create abstract conceptual structures [5, 31, 33–35]. However, these assumptions do not address the relational and unstable nature of preconceptual structures such as basic-level categories: for example, the perception and understanding of categories and category boundaries can differ substantially across individuals, situations and cultures and even change dramatically over time. Haser [36] argues that Lakoff’s account of basic-level categories does not capture the ‘kinds of meaning differences that are crucial to semantics’ [36, p. 136]. Jolicoeur et al. [37] suggested that the ‘level at which objects are identified first depends on typicality’ [37, p. 271], and the factors responsible for an individual’s preference for a basic-level term ‘may include not only those associated with its basic-level category but also the larger set of background causes that govern the individual’s environment’ [38, p. 9]. These factors may shift the entry level of categorization to the subordinate level [4]. Jolicoeur et al. argued that the notion of entry point must be understood as an ‘attribute of individual exemplars rather than an attribute of categories’ [37, p. 272]. In this regard, Anaki and Bentin [39] found that familiar towers (e.g. Leaning Tower of Pisa) are categorized faster at the instance level than at the basic level. Since the categorization of familiar stimuli was slower and less accurate at the basic and superordinate levels than unfamiliar stimuli, they suggested that, regardless of expertise, detailed semantic information may stimulate a shift of the entry level of categorization from the basic to the instance (individual) level. Belke et al. [40] also observed that art objects were most frequently categorized at the subordinate level when compared with non-art objects. For example, expert viewers were able to access artist-specific representations as quickly as basiclevel representations of everyday objects: a painting by Picasso was identified as an instance of a Picasso within the same time frame as a depiction of a car was identified as an instance of a car [39, p. 200]. Belke et al. suggested that the ‘artist’s name has a special status in the memorial representation of visual art and serves as a predominant entry point in recognition in art perception’ [40, p. 191]. Even though a number of studies have addressed the preference of children for basic-level categories [22, 41, 42], Mandler and her associates [43–45] found no differentiation for basic-level categories. She argued that [46, p. 294]: [t]he data we have presented suggest that in the domains of animals, plants, and vehicles, and perhaps in other domains as well, basic-level categories, as formulated by Rosch and Mervis (1975) and Rosch et al. (1976), do not form the entry level in the development of hierarchical categorical systems. Rather, children appear to begin categorizing at a more global level; only gradually do these global categories become differentiated.

In an attempt to generalize her findings, Mandler argued that ‘various differences between perceptual and conceptual categories illustrate the insufficiency of the widely held view that ‘‘basic-level’’ categories are the first to be formed’ [47, p. 289]. She described the hypothesis that basic-level categories were the first to be leaned as incorrect because, according to her studies, infants tended to form categories at the superordinate level. Furthermore, she claimed that Rosch’s experimental studies did not ‘constitute adequate tests of the hypothesis that basic-level concepts are the first to be formed, nor did they work with young enough subjects’ [47, p. 289]. Mandler speculated that Rosch’s experiments using sorting applied a too strict criterion, mandating that sorting was judged to be ‘incorrect if a superordinate class was subdivided, even though the subjects were not told how many divisions to make’ [47, p. 289]: for example, placing shoes and socks in one category and shirts and pants in another category would be considered incorrect. Mandler argued that, to test whether categorization at the basic level was easier than at the superordinate level, it was not appropriate to use the sorting of dogs and cars as the basic level and to contrast those with the sorting of animals and vehicles: even though dogs and cars are basic-level categories, they entail what she called the ‘superordinate contrast’ [47, p. 289]. She cautioned that such extreme contrasts should be avoided when testing for basic-level categorization and suggested that contrasting pairs should involve entities in the same hierarchy (e.g. dogs and rabbits or cars and motorcycles). Journal of Information Science, 39 (5) 2013, pp. 676–687 Ó The Author(s), DOI: 10.1177/0165551513481443

Hajibayova

680

Murphy [13] has pointed out that Mandler’s attempts to cast doubt on the theory of basic-level categories have not been widely supported. He has argued that her criticism that basic-level theory was based on the assumption that the ‘basic level is an immutable property of categories that is predictable from the ‘‘objective category’’ itself’ [13, p. 235] would, in fact, be falsified by identification of different basic-level categories across either cultures or developmental stages. Moreover, Murphy also contended that, because the notion of basic level categories is a ‘psychological construct, . [it] depends on what people know about the categories . [and] can only be identified by behavioral tests . [T]o be compelling, such criticism needs to be backed by alternative, concrete proposals for the phenomena Rosch, Mervis and their colleagues discovered’ [13, p. 235].

3.2. Operationalization and explanation of basic-level categories Rosch [10] and her colleagues formalized the basic-level categories in terms of the probability of cue validity. She operationalized the cue validity of a category as the sum of the individual cue validities for each attribute of a category, because a category with high cue validity would be more distinct from other categories than a category with low cue validity. Rosch postulated that basic objects (such as chair and car) were at the most inclusive level where attributes were common to all or most members of the basic category, and suggested that ‘both total cue validities and category resemblance are maximized at that level of abstraction at which basic objects are categorized’ [10, p. 31]. In comparison with basic-level categories, superordinate categories possess fewer common attributes and, thus, demonstrate lower total cue validity and category resemblance. Subordinate categories also have lower cue validity than do basic-level categories, because they share many of their attributes with the other subordinate categories of the basic-level category such that the ‘weight of the added common features tend[s] to exceed the weight of the distinctive features’ [10, p. 31]. There have been several studies suggesting alternative methods for formally defining basic-level categories. Jones [48] proposed his so-called category-feature collocation metric calculated as the product of the probabilities that a possessor of the feature is a member of the category and that a member of the category is a possessor of the feature. Corter and Gluck [49] developed the category utility metric to predict the preferred level of categorization using subject reaction time to verify and name pictures. One of the major problems with all of these metrics is that they are based on frequency of feature occurrence, which complicates the specification of features [50]. For instance, to define the cue validity of DOG, it is necessary to decide which features should be included (e.g. ‘BARKS’, ‘CAN SERVE AS A GUARD’). Features should be prioritized to assure that category instances in the environment are represented. Murphy has argued that defining which ‘features are really in the environment has not been adequately addressed by metrics of the basic level’ [13, p. 216] and has suggested that performance measures may overcome the shortcomings of cue validity, category feature collocation and category utility metrics. Several studies have attempted to provide a theoretical explanation for how basic-level categories are perceived and processed. Murphy [13, 51] has suggested a differentiation explanation for the preference for basic-level categories. Murphy and Brownell [51] proposed that the distinctiveness and specificity of the category determine category differentiation: specificity refers to how specific and informative the category is, and distinctiveness refers to how dissimilar the categories are in comparison to contrast categories. In a series of experiments with natural language categories, they found that objects could be identified faster when they were members of differentiated categories. In particular, an atypical subordinate category (e.g. RACING CAR) was highly differentiated and, therefore, was responded to as quickly as a basic-level category during object recognition. Murphy and Brownell argued that these two factors usually oppose each other: ‘[V]ery specific categories tend to be very similar to their contrast categories, which differ in only one or two features; very general categories tend to be distinctive’ [51, p. 71]. They also suggested that categories at the basic level are most differentiated because they have relatively positive values for both factors rather than a maximally strong value for only one. Murphy [13] subsequently suggested that basic categories were most easily differentiated in terms of informativeness and distinctiveness. The informativeness of a basic-level category refers to the fact that the category communicates a large amount of information: when you know the basic-level category to which an object belongs, you can infer significant information about that object. The distinctiveness of a basic-level category refers to how different a category is when contrasted with coordinate categories at the same level in the hierarchy. According to the differentiation explanation, superordinate categories are distinctive, but less informative than basic-level categories, whereas subordinate categories are more informative but significantly less distinctive than basic-level categories. Explanations of basic-level categories based on differentiation have been tested by several experimental studies [52]. Murphy and Smith [24] argued that distinctive perceptual attributes underlie the experimental finding that objects are categorized fastest at the basic level. To test their hypothesis, they asked subjects to learn the hierarchical structure of 14 categories of tools: two superordinate categories, four basic-level categories and eight subordinate categories. Timed Journal of Information Science, 39 (5) 2013, pp. 676–687 Ó The Author(s), DOI: 10.1177/0165551513481443

Hajibayova

681

categorization of objects using the learned categories demonstrated that basic-level categories were identified fastest, followed by subordinate and superordinate categories, respectively. Rogers and Patterson [53] have proposed an explanation of basic-level categories that is based on the theory of parallel distributed processing (PDP) of conceptual knowledge. PDP theory is grounded on the notion that knowledge of words and objects follows from interaction of various representations, including perceptual and motor as well as linguistic features. According to PDP theory, the ‘ability to interpret and make use of new information depends on knowledge that has previously accumulated in the semantic network through gradual learning; but neither the information nor its interpretation is stored immediately in the network’s connection weights’ [53, p. 63]. For example, the category COFFEE is associated with various features, such as taste, flavour and smell, from which the definition of the category emerges. Rogers and Patterson [53] adopted the proposal of Barsalou et al. [54] that semantic memory consists of a set of representations associated with a ‘hub’ that facilitates interactions between perceptual, motor and language representations. Thus, the semantic representation of a category would be a hub that allowed the ‘instantiation of a representation in one modality to provoke the instantiation of associated words, actions, and perceptual characteristics in other modalities’ [53, p. 457]. For example, if it is learned that an instance of PEAR HAS SEEDS, this acquired information would be generalized to other members of the category FRUIT; however, this generalization would not apply to a LIGHT BULB, regardless of similar shape, because LIGHT BULB represents a different pattern in the hub. Rogers and Patterson emphasized that the ‘similarity structure in the hub comes to mirror the similarities apparent across all of the different sensory, motor, and language representations to which it is connected – so that items that have many different kinds of features in common (similar shapes, movements, functions, colors, and so on) come to be represented as similar’ [53, p. 458]. It follows that the basic level occurs owing to similarity of the structure of representations in the hub. For example, while members of the categories ORANGE and BALL share a similar shape, they vary in terms of use, purpose, etc. However, although members of the categories ORANGE and STRAWBERRY differ in colour and shape, they share many common features. This explanation differs from the differentiation theory in that the emergence of the basic level ‘depends upon the pattern of generalization fostered by these representations as the network learns to name’ [53, p. 459]. In other words, a learned item is generalized to items with similar representations. For example, learning that a canary is a type of BIRD associates it with other types in the category. Rogers and Patterson argued that ‘similarity-based generalization . promotes more rapid learning of basic-level names for words of equal frequency’ [53, p. 460]. On the other hand, they speculated that the entry point level may be shifted depending upon the ‘exact similarity structure of the semantic representations and the frequency of the labels at different levels of specificity’ [53, p. 460]. In contrast to the theory of differentiation, PDP theory can account for time of activation at different levels of categories. Superordinate categories are activated sooner than basic-level categories because they ‘apply broadly across a wide range of semantically related items’ [53, p. 461]. However, because of the broad application of superordinate categories, full activation (i.e. sufficient to provide response) will be slower: ‘[B]roadly spaced representations to which . [superordinate categories] apply do not foster much generalization one to another so the name is slower to activate’ [53, p. 461]. In contrast, because basic-level categories represent a relatively narrower space, their activation starts later but full activation is achieved more rapidly. Thus, the predicted time of activation starts with the superordinate category (ANIMAL) and moves to the basic category (BIRD) and then to the subordinate category (CANARY). Theorizing that the time to define category membership ‘depends upon one of the name outputs exceeding a threshold’ [53, p. 461], Rogers and Patterson have suggested that, in the case of a high threshold for responding, the basic-level category will be verified first. However, when there is a need to respond at increasingly shorter latencies, the ‘basic-over . [superordinate] level advantage should first vanish and then be reversed – so that, at very fast latencies, people should be more accurate for superordinate relative to basic categorization’ [53, p. 461]. However, it appears that the relational nature of basic-level categories, which leads to problems in applying basiclevel category theory across individual(s), situations and cultures, has not yet been vigorously studied. In this regard, Heidegger’s [55] notion of ‘handiness’ or Zuhandenheit frames our understanding of relational nature of basic-level categories and offers critical insights for more effective operationalization of basic-level categories.

3.3. Basic-level categories through the lens of Heidegger’s notion of handiness Heidegger defines notion of ‘at hand’ or ‘handiness’ as the ‘ontological categorical definition of beings as they are ‘‘in themselves’’’ [55, p. 67]; representation of being in a relation to useful things that are handy on the ‘basis of what is objectively present’ [55, p. 67]. For Heidegger, a useful thing is ‘essentially ‘‘something in order to .’’ . such as serviceability, helpfulness, usability, handiness, [that] constitute[s] a totality of useful things’ [55, p. 64]. He argued that Journal of Information Science, 39 (5) 2013, pp. 676–687 Ó The Author(s), DOI: 10.1177/0165551513481443

Hajibayova

682

the structure of ‘in order to’ includes a ‘reference of something to something’; in particular, useful things ‘always are in terms of their belonging to other useful things’ [55, p. 64]: writing materials, pen, ink, paper, desk blotter, table, lamp, furniture, windows, doors, room[:] These ‘things’ never show themselves initially by themselves, in order then to fill out a room as a sum of real things. What we encounter as nearest to us, although we do not grasp it thematically, is the room, not as what is ‘between the four walls’ in a geometrical, spatial sense, but rather as material for living.

However, handiness cannot be grasped or explained theoretically, but by an association that ‘makes use of things’, ‘guides our operations and gives them their specific thingly quality’ [55, p. 65]. For example, hammering is associated with a hammer, whereas the act of hammering itself ‘discovers the specific ‘‘handiness’’ of the hammer’ [55, p. 65]. In other words, handiness of a useful thing is discovered in association with the ‘work to be produced’ [55, p. 65]: The shoe to be produced is for wearing (footgear), the clock is made for telling time. The work which we primarily encounter when we deal with things and take care of them – what we are at work with – always already lets us encounter the what-for of its usability which essentially belongs to it.

In Heidegger’s [55] terms basic-level categories are ‘useful things’ that ‘reveal’ themselves by their handiness. Moreover, it seems that a ‘thingly’ [55, p. 65] quality of basic-level categories underlies the predominance of basic-level categories in our everyday vocabulary. One may speculate that it is probably due primarily to a general understanding of ‘work to be produced’ [55, p. 65]. However, Heidegger sees the work produced not only as a reference to the ‘what-for of its usability’ and the ‘whereof of which it consists’ [55, p. 66], but also to the ‘surrounding world of nature’ [55, p. 67]. He exemplifies the use of the clock, arguing that: When we look at the clock, we tacitly use the ‘position of the sun’ according to which the official astronomical regulation of time is carried out. The surrounding world of nature is also at hand in the usage of clock equipment which is at first inconspicuously at hand. [55, p. 67]

For Heidegger, the handiness of things in the ‘work world nearest to us’ [55, p. 67] is based on the function of ‘discovering’ and depends upon the ‘way we are absorbed[;] innerworldly beings that are brought along together with their constitutive references are discoverable in varying degrees of explicitness and with a varying [degree of] attentive penetration’ [55, p. 67]. In other words, objects or ‘things’ are utilized and used in relation to our perception and understanding of ‘things’ and as such they do not have their own independent meaning. Apparently, basic-level categories vary across individuals and cultures because of variations in ‘innerworldly beings’, and particularly in their relations to or their understanding of the ‘thingly quality’ [55, p. 65] of objects. For Heidegger, variation is phenomenological in that it comes out of our contextualized experience of engaging with objects. Through the lens of Heidegger’s notion of handiness, it seems that basic-level categories should be defined within individuals’ contextualized experience of using objects or an understanding of their ‘thingly quality’ [55, p. 65].

4. LIS studies of the basic-level categories Jacob [56] has argued that cognitive categorization is the baseline for formal systems of organization but that the ‘organization imposed on cognitive categories is so dynamic and responsive to changes in context that it cannot establish persistent, knowledge-bearing relationships between categories’ [56, p. 533]. Conventional knowledge organization systems, such as the Dewey Classification System, are rigid structures of information representation that cannot account for the fuzzy cognitive aspect of how the world in general and information in particular are perceived and organized by individuals. It is not surprising that theories of categorization – particularly, the theory of basic-level categories – are rarely employed in library and information science (LIS). Overall, LIS research has tended to confine investigation of basiclevel categories to two questions: how the basic level of categories is represented in knowledge organization systems [6, 57, 58] and how basic-level terms are used to represent the conceptual content of resources and, especially, images [7, 59–61]. For example, Fernandez and Eastman [58] had looked at the relationship between levels of specificity and the Medical Subject Heading (MeSH) hierarchical tree structure for the term endocrine disease. The study observed that the highest number of terms and postings occurred at the middle level of the MeSH hierarchy. Fernandez and Eastman suggested that this occurred because most terms at the intermediate level actually corresponded to ‘fundamental Journal of Information Science, 39 (5) 2013, pp. 676–687 Ó The Author(s), DOI: 10.1177/0165551513481443

Hajibayova

683

concepts with respect to endocrine diseases’ [58, p. 63]. They argued that this finding appeared to contradict the commonly held belief that specific terms were used more often to represent the conceptual content of a resource. It also called into question Iyer’s [57] argument that basic-level terms fail to represent the specificity of categories and therefore make poor search terms in information retrieval systems. However, Fernandez and Eastman do not provide any justification for equating the middle level of the MeSH hierarchy with basic-level categories. In a series of influential studies of image categorization, Jo¨rgensen [7, 62] adopted Rosch’s research on basic-level categories to study the indexing and retrieval of images of perceptual class objects (e.g. clothing, body parts). Summarizing a series of studies on image retrieval, Jo¨rgensen [7] reported that participants in viewing, searching and memory tasks generally employed basic-level terms (76.6, 77.9 and 73.5%, respectively). When compared with the results of other tasks, participants in the memory task employed fewer basic-level terms and slightly more superordinate terms (i.e. 14.6% as opposed to 11.8 and 11.7% for viewing and searching tasks, respectively). These findings could be interpreted as support for earlier studies indicating that objects are generally recognized at the basic level [4]; however, Jo¨rgensen does not provide specifics as to how the different levels of categories were operationalized. The most extensive study of basic-level categories reported in the LIS literature was conducted by Green and her colleagues [6, 63]. Green et al. [63] initially investigated the occurrence of basic-level categories in four different systems of knowledge organization: a bilingual thesaurus (Canadian Literacy Thesaurus), a biomedical vocabulary (Unified Medical Language System) and two ontologies (ThoughtTreasure and WordNet). Analysis of a random sample that included the semantic scope of a term as well as its placement in the original hierarchy demonstrated that equivalence across systems occurred significantly at the basic level rather than at subordinate or superordinate levels. Interestingly, basic-level terms were observed most often in the Canadian Literacy Thesaurus and least often in the Unified Medical Language System. Analysis using goodness-of-fit tests confirmed that, in these four systems, basic-level terms were more likely to have equivalents across systems than subordinate or superordinate terms. In a subsequent study, Green [6] investigated the universality of basic-level categories in a sample of 28 thesauri from 10 subject domains based on a set of basic-level terms collected from WordNet, a hierarchically structured lexical database of English words. Based on analysis of WordNet representation of the six nonbiological categories (MUSICAL INSTRUMENT, FRUIT, TOOL, CLOTHING, FURNITURE and VEHICLE) used in Rosch et al.’s [4] study, Green [6, p. 8] identified six general criteria for the identification of basic-level categories: • • • • • •

length and structure of lexical units (i.e. whether the lexical unit is a simple word, a compound word or a phrase); level of occurrence within the WordNet tree structure; number of links to parts; frequency of usage; total number of links to other categories; and number of children/subordinate categories, both immediate (one level down) and overall (in all subordinate).

Green applied these criteria to the WordNet noun network, analysing every leaf node in WordNet’s 59,692 hierarchies, and identified 7,168 basic-level categories. To investigate the hypothesis that categories at the basic level are more universal and therefore more likely to cooccur across classificatory systems than either subordinate or superordinate categories, Green identified online thesauri for 11 subject domains (i.e. agriculture, education, engineering, environment, graphic materials, health, information science, legislation, political science, population science and water sciences) and randomly selected 10 category terms from each thesauri. The selected category terms were then expanded using the relational structure of the relevant thesaurus. For each category term in these hierarchical expansions, the closest corresponding category term was identified in each of the other thesauri for that subject domain to generate category pairs consisting of one category term from the hierarchical expansion in one thesaurus and the closest corresponding term in a second thesaurus. These category pairs were then analysed to determine if the two category terms were equivalent, and basic-level category terms were identified based on the criteria for basic-level categories generated from the analysis of WordNet. Green’s findings appeared to support her hypothesis of the universality of basic-level terms. She suggested that the practical benefit of the study was that it provided direction for building crosswalks between classificatory systems and argued that, if basic-level categories were significantly more likely to have exact equivalents across systems than categories not at the basic level, then developing mappings across classificatory systems should emphasize basic-level categories to generate the ‘cleanest mappings’ [6, p. 8]. However, even though the universality of basic-level categories in systems of knowledge organization and representation would seem to indicate their universality in indexing as well, basic-level categories may not be appropriate for supporting collocation precisely because they have a high probability of occurrence. Journal of Information Science, 39 (5) 2013, pp. 676–687 Ó The Author(s), DOI: 10.1177/0165551513481443

Hajibayova

684

Green’s [6] criteria for operationalizing different levels of categorization have been adapted in studies of image and emotion categorizations [8, 59, 61]. For example, Rorissa and Iyer [8], in two separate studies of prototypes and preferred levels of categorization in image categorization, observed predominance of superordinate categories (54.17 and 58.88%), followed by basic-level categories (42.5 and 36.42%) and subordinate categories (3.33 and 4.70%). In a subsequent study, Rorissa [59] examined category terms assigned to individual images and to groups of images. In line with the earlier research of Rorissa and Iyer [8], categories assigned to groups of images tended to be at the superordinate level, and categories assigned to individual images displayed a preference for basic-level categories (63.29%) over subordinatelevel (27.78%) and superordinate-level (8.93%) categories. Schmidt and Stock [61] also applied the coding scheme developed by Green [6] and used by Rorissa and Iyer [8, 59] to analyse and identify levels of categorization commonly used to represent emotion in images. Subjects were asked to express their perceptions of the intensity of five fundamental emotions (i.e. anger, disgust, fear, happiness and sadness), and provide category terms representing the content as well as impressions and associations evoked by 30 Flickr1 images. The study suggested that participants tended to provide basic-level categories to express emotions (e.g. HAPPY rather than its subordinates EUPHORIC or CHEERFUL). However, because WordNet does not include adjectives, Schmidt and Stock [61] had to identify derivative nouns such as happiness and happy to identify hierarchical relationships. Several studies of online social tagging vocabulary have suggested that taggers tend to use basic-level terms as bookmarks [64–67]. In a seminal study of Delicious2 tags, Golder and Huberman speculated that ‘earlier tags in a bookmark represent basic levels, because they are not only widespread in agreement, but are also the first terms that users thought of when tagging the URLs in question’ [64, p. 202]. Taking into consideration the fact that basic-level terms may vary across users, they observed that, ‘for the purposes of tagging systems, however, conflicting basic levels can prove disastrous, as documents tagged perl and javascript may be too specific for some users, while a document tagged programming may be too general for others’ [64, p. 200]. However, their comments are little more than assumptions based on earlier experimental studies [20] rather than empirical observation. Golder and Huberman have also suggested that the supposed stabilization of tags assigned to Delicious’s two most popular URLs might have been due to user agreement on vocabulary representing the basic level of categorization, an hypothesis that appears to be consonant with Noruzi’s [65] suggestion that the ‘basic level, to superordinate (more general) and subordinate (more specific) levels, is that which is most directly related to humans’ interactions’ [65, p. 203]. In their analysis of Delicious tags, Munk and Mork defined basic-level terms as the ‘most popular keywords . [that are] very general and may be said to belong to a number of cognitive basic categories which many people working professionally in the IT field will consider if asked to mention some IT terms off the top of their heads’ [67, p. 121] They claimed that the 245 most frequently used tags in their study, or 37.2% of their sample, were basic-level categories and argued that, in a cognitive sense, these tags could be viewed as the ‘shortest possible and most economical encoding[s] of a relevant category’ that ‘everyone could think of without giving it much thought’ [67, p. 121]. They concluded that the most popular tags used on Delicious represented the basic level of categorization: ‘[B]road basic cultural content categories are predominant (such as: music, news, design), while broad basic categories for everyone working professionally with IT are also frequent (such as: programming, Linux, javascript, flash)’ [67, p. 121]. In light of previous studies suggesting that domain knowledge may shift the entry level from the basic level of categorization to a subordinate level [37, 20], the supposed basic-level terms used by IT professionals to tag resources on Delicious appear to be predominantly subordinate terms. Moreover, Munk and Mork’s reference to the ‘broad basic cultural content categories’ is also questionable because no justification for characterization of the terms ‘music’, ‘news’ and ‘design’ as basic-level categories was provided. Despite the claim that their use of basic-level category theory was based on Rosch et al.’s work [3, 4], Munk and Mork’s conclusions are questionable: they provided neither a clear definition of basic-level categories nor an explanation of how basic-level categories were operationalized in their research. In sum, Munk and Mork’s application of the theory of basic-level categories seems to be more intuitive than based on evidence. Although the universality of basic-level terms has been recognized and positive implications for information representation and organization have been suggested [6, 68], few studies have actually tested their validity in the LIS environment, and too many of these studies do not provide an account of how the basic-level categories were defined and operationalized. Several mainstream studies [8, 59, 61] have adapted the coding scheme that Green [6] devised based on WordNet. WordNet is a hierarchically structured lexical database of English words that has become the ‘de facto standard because it is freely available for research’ [69, p. 52]. Unfortunately, however, WordNet was built on paradigmatic relations between nouns and verbs, which may lead to difficulties when attempting to determine the exact semantic level of a term. Problems associated with operationalization of basic-level terms based on the hierarchical structure of synsets in WordNet are not limited to the determination of a term’s exact semantic level. More importantly, operationalization of basic-level terms based on the WordNet hierarchical structure of synsets masks the importance of the context in which a category term occurs and thus fails to take into account the relational nature – the ‘handiness’ [55] – of basic-level Journal of Information Science, 39 (5) 2013, pp. 676–687 Ó The Author(s), DOI: 10.1177/0165551513481443

Hajibayova

685

terms. The implications of Heidegger’s [55] notion of handiness for basic-level category research in LIS is obvious: the operationalization of basic-level categories based on the hierarchical structure of normalized vocabularies cannot account for variations in the relational structures of individuals, much less variations in shared conceptual structures across cultures. A more effective approach to operationalizing basic-level categories must take into account the contextualized experiences of individuals actively engaging with objects in order to assess their understanding of an object’s ‘thingly quality’ [55, p. 65] and its relationship to other objects.

5. Conclusions Basic-level categories vary across individuals and cultures because of differences in the everyday experiences and activities of individuals. Within the framework of Heidegger’s [55] notion of handiness, variation is both natural and phenomenological in that perception and understanding – and thus the meaning of ‘things’ – arise out of the individual’s contextualized experiences of engaging with objects. Therefore, an effective approach to operationalizing basic-level categories should be based on individuals contextualized experiences of engaging with objects, their understanding of an object’s ‘thingly quality’ (p. 65) and its relationship to other objects in the everyday world of work. Moreover, owing to the multidisciplinary nature of the basic-level categories research, a cross-disciplinary approach is needed to study basic-level categories. Cross-disciplinary studies of basic-level categories may help not only to better understand individuals categorization patterns, but also to contribute to development of systems that would provide more ‘organic’ juxtaposition of controlled or ‘artificial’ [70] and user-generated vocabularies. As a result, it may improve representation, organization, ‘findability’ [71] and effective use of the knowledge. Acknowledgment I express my sincere thanks and appreciation to Dr Elin Jacob for generous help with this study. I also would like to thank Dr Mary Gray for valuable suggestions and reviewers for their useful suggestions.

Notes 1. Flickr, http://www.flickr.com 2. Delicious, http://delicious.com

References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18]

Brown R. How shall a thing be called? Psychological Review 1958; 65(1): 14–21. Wittgenstein L. Philosophical investigations. Oxford: Blackwell, 1963. Rosch E and Mervis C. Family resemblances: Studies in the internal structure of categories. Cognitive Psychology 1975; 7(4): 573–605. Rosch E, Mervis C, Gray W, et al. Basic objects in natural categories, Cognitive Psychology 1976; 8(3): 382–439. Lakoff G. Women, fire, and dangerous things: What categories reveal about the mind. Chicago, IL: The University of Chicago Press, 1987. Green R. Vocabulary alignment via basic level concepts, http://www.oclc.org/research/grants/reports/green/rg2005.pdf (2006, accessed June 2012). Jo¨rgensen C. Image retrieval: Theory and research. Lanham, MD: Scarecrow Press, 2003. Rorissa A and Iyer H. Theories of cognition and image categorization: What category labels reveal about the basic level theory. Journal of the American Society for Information Science and Technology 2008; 59(9): 1383–1392. Zerubavel E. The fine line: Making distinctions in everyday life. Chicago, IL: University of Chicago Press, 1991. Rosch E. Principles of categorization. In: Rosch E and Lloyd B (eds) Cognition and categorization. Hillsdale, NJ: Lawrence Erlbaum, 1978. Smith E and Medin D. Categories and concepts. Cambridge, MA: Harvard University Press, 1981. Bruner J, Goodnow J and Austin G. The process of concept attainment. In: Margolis E and Laurence S (eds) Concepts: Core readings. Cambridge, MA: The MIT Press, 1999. Murphy G. The big book of concepts. Cambridge, MA: The MIT Press, 2002. Rosch E. Natural categories. Cognitive Psychology 1973; 4(3): 328–350. Heider E. ‘Focal’ color areas and the development of color names. Developmental Psychology 1971; 4(3): 447–455. Heider E. Universals in color naming and memory. Journal of Experimental Psychology 1972; 93(1): 10–20. Medin D, Wattenmaker W and Hampson S. Family resemblance, conceptual cohesiveness, and category construction. Cognitive Psychology 1987; 19: 242–279. Smith E. Concepts and induction. In: Posner M. (ed.) Foundation of cognitive science. Cambridge, MA: The MIT Press, 1990.

Journal of Information Science, 39 (5) 2013, pp. 676–687 Ó The Author(s), DOI: 10.1177/0165551513481443

Hajibayova [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52]

[53]

686

Laurence S and Margolis E. Concepts and cognitive science. In: Margolis E and Laurence S (eds) Concepts: Core readings. Cambridge, MA: The MIT Press, 1999. Tanaka J and Taylor M. Object categories and expertise: Is the basic level in the eye of the beholder? Cognitive Psychology 1991; 23(3): 457–482. Markman A and Wisniewski E. Similar and different: The differentiation of basic-level categories. Journal of Experimental Psychology: Learning, Memory, and Cognition 1997; 23 (1): 54–70. Blewitt P. Dog versus collie: Vocabulary in speech to young children. Developmental Psychology 1983; 19(4): 602–609. Newport E and Bellugi U. Linguistic expression of category levels in a visual–gestural language: A flower is a flower is a flower. In: Rosch E and Lloyd B (eds) Cognition and categorization. Hillsdale, NJ: Lawrence Erlbaum, 1978. Murphy G and Smith E. Basic-level superiority in picture categorization. Journal of Verbal Learning and Verbal Behavior 1982; 21(1): 1–20. Mack M, Wong A, Gauthier I, et al. Time course of visual object categorization: Fastest does not necessarily mean first. Vision Research 2009; 49(15): 1961–1968. Tversky B and Hemenway K. Categories of environmental scenes. Cognitive Psychology 1983; 15(1): 121–149. Tversky B and Hemenway K. Objects, parts, and categories. Journal of Experimental Psychology: General 1984; 113(2): 169– 193. Berlin B and Kay P. Basic color terms: Their universality and evolution. Berkeley, CA: University of California Press, 1991. Berlin B, Breedlove D and Raven P. Principles of Tzeltal plant classification. New York: Academic, 1974. Berlin B. Ethnobiological classification. In: Rosch E and Lloyd B (eds) Cognition and categorization. Hillsdale, NJ: Lawrence Erlbaum, 1978. Lakoff G and Johnson M. The metaphorical structure of the human conceptual system. Cognitive Science 1980; 4: 195–208. Lakoff G and Johnson M. Philosophy in the flesh: The embodied mind and its challenge to western thought. New York: Basic Books, 1999. Lakoff G and Johnson M. Metaphors we lve by. Chicago, IL: University of Chicago Press, 1980. Ko¨vecses Z. Conceptual methaphor theory: Some criticisms and alternative proposals. Annual Review of Cognitive Linguistics 2008; 6: 168–184. Giora R. Is metaphor unique? The Cambridge handbook of metaphor and thought. In: Gibbs R (ed.) The Cambridge handbook of metaphor and thought. New York: Cambridge University Press, 2008. Haser V. Metaphor, metonymy, and experientialist philosophy: Challenging cognitive semantics. Berlin: Walter de Gruyter GmbH, 2005. Jolicoeur P, Gluck M and Kosslyn S. Pictures and names: Making the connection. Cognitive Psychology 1984; 16(2): 243–275. Rips L, Blok S and Newman G. Tracing the identity of objects. Psychological Review 2006; 113(1): 1–30. Anaki D and Bentin S. Familiarity effects on categorization levels of faces and objects. Cognition 2009; 111(1): 144–149. Belke B, Leder H, Harsanyi G, et al. When a Picasso is a ‘Picasso’: The entry point in the identification of visual art. Acta Psychologica 2010; 133(2): 191–202. Greco C and Daehler M. Immediate and long-term retention of basic-level categories in 24-month-olds. Infant Behavior and Development 1985; 8: 459–474. Mervis C and Crisafi M. Order of acquisition of subordinate-, basic-, and superordinate-level categories. Child Development 1982; 53(1): 258–266. Mandler J and Bauer P. Cradle of categorization: Is the basic level basic? Cognitive Development 1988; 3: 247–264. Mandler J, Fivush R and Reznick S. The development of contextual categories. Cognitive Development 1987; 2: 339–354. Mandler J and McDonough L. Concept formation in infancy. Cognitive Development 1993; 8: 291–318. Mandler J, Bauer P and McDonough L. Separating the sheep from the goats: Differentiating global categories. Cognitive Psychology 23 (1991). Mandler J. Representation. In Damon W (ed.) Handbook of child psychology: Cognition, perception, and language. Hoboken, NJ: John Wiley & Sons, 1998. Jones G. Identifying basic categories. Psychological Bulletin 1983; 94(3): 423–428. Corter J and Gluck M. Explaining basic categories: Feature predictability and information. Psychological Bulletin 1992; 111(2): 291–303. Murphy G and Lassaline M. Hierarchical structure in concepts and the basic level of categorization. In: Lamberts K and Shanks D (eds) Knowledge, concepts and categories. London: UCL Press, 1997. Murphy G and Brownell H. Category differentiation in object recognition: Typicality constraints on the basic category advantage. Journal of Experimental Psychology: Learning, Memory, and Cognition 1985; 11(1): 70–84. Lassaline M, Wisniewski E and Medin D. The basic level in artificial and natural categories: Are all basic levels created equal? In: Burns B (ed.) Percept, concepts, and categories: The representation and processing of information. Amsterdam: Elsevier, 1992. Rogers T and Patterson K. Object categorization: Reversals and explanations of the basic-level advantage. Journal of Experimental Psychology: General 2007; 136(3): 451–469.

Journal of Information Science, 39 (5) 2013, pp. 676–687 Ó The Author(s), DOI: 10.1177/0165551513481443

Hajibayova [54] [55] [56] [57] [58]

[59] [60] [61] [62] [63]

[64] [65] [66] [67] [68] [69] [70] [71]

687

Barsalou L, Simmons W, Barbey A, et al. Grounding conceptual knowledge in modality-specific systems, Trends in Cognitive Sciences 2003; 7: 84–91. Heidegger M. Being and time: A translation of Sein and Zeit, translated by Joan Stambaugh. SUNY Series in Contemporary Continental Philosophy, Schmidt DJ (ed.). Albany, NY: State University of New York Press, 1996. Jacob E. Classification and categorization: A difference that makes a difference. Library Trends 2004; 52(3): 515–540. Iyer H. Classificatory structures: Concepts, relations and representation. Frankfurt: Indeks, 1995. Fernandez M and Eastman C. Basic taxonomic structures and levels of abstraction. Advances in classification research. In: Humphrey S and Kwasnik B (eds) Proceedings of the 1st ASIS SIG/CR classification research workshop, Vol. 1. Medford, NJ: Learned Information, 1990. Rorissa A. User-generated descriptors of individual images versus labels of groups of images: A comparison using basic level theory. Information Processing and Management 2008; 44: 1741–1753. Yoon J. Towards a user-oriented thesaurus for non-domain-specific image collection, Information Processing and Management 2009; 45: 452–468. Schmidt S and Stock W. Collective indexing of emotions in images. A study in emotional information retrieval. Journal of the American Society for Information Science and Technology 2009; 60(3): 863–876. Jo¨rgensen C. Image attributes: An investigation. Unpublished doctoral dissertation. Syracuse: Syracuse University, 1995. Green R, Bean C and Hudon M. Universality and basic level concepts. In: Lo´pez-Huertas M (ed.) Challenges in knowledge representation and organization for the 21st century: Integration of knowledge across boundaries: Proceedings of the seventh international ISKO conference, Vol. 8. Wu¨rzburg: Ergon, 2002. Golder S and Huberman B. Usage patterns of collaborative tagging systems. Journal of Information Science 2006; 32(2): 198– 208. Noruzi A. Folksonomies: (Un)controlled vocabulary? Knowledge Organization 2006; 33(4): 199–203. Munk T and Mork K. Folksonomy, the power law & significance of the least effort. Knowledge Organization 2007; 34 (1): 16– 33. Munk T and Mork K. Folksonomies, tagging communities, and tagging strategies – An empirical study. Knowledge Organization 2007; 34(3): 115–127. Bates M. Indexing and access for digital libraries and the Internet: Human, databases, and domain factors. Journal of the American Society for Information Science 1998; 49(13): 1185–1205. Ide N and Wilks Y. Making sense about sense. In: Agirre E and Edmonds P (eds) Word sense disambiguation: Algorithms and applications. Berlin: Springer, 2007. Lancaster F. Vocabulary control in information retrieval systems. Advances in Librarianship 1970; 7: 1–40. Morville P. Ambient findability. Sebastopol: O’Reilly Media, 2005.

Journal of Information Science, 39 (5) 2013, pp. 676–687 Ó The Author(s), DOI: 10.1177/0165551513481443