Synthetic classification of museum artifacts using basic concepts

Rick Szostak, University of Alberta, Canada

Abstract

Keywords: Classification, Synthetic, Search; Basic Concepts

1. Introduction

The purpose of this paper is to argue that it is entirely feasible to provide subject access to museum artifacts (and also art gallery artifacts and archival documents). The paper then outlines how this could be done in practice.

It should be stressed that this paper argues for an approach based on subject classification. Many in the museum community may prefer an approach that involves the searching of object descriptions for keywords. This would mean that museums only need provide online access to existing object descriptions. The unfortunate reality—demonstrated in both library and online searches—is that keyword searching misses much that a user seeks due to the simple fact that different terminology is employed for the same thing in different databases. This problem is likely particularly acute for museums where similar artifacts from different cultures are often accorded different names and descriptions.

The alternative to keyword searching is subject searching using controlled vocabulary (that is, where only certain search terms are allowed). This is generally found by scholars to be more precise: it allows them to find particular works in libraries or online that better meet their needs than they could find through keyword searching alone. But museums rightly shy away from adopting the needlessly complex systems of subject classification employed in the world’s libraries. Indeed, few online databases of any type have shown any proclivity for library classifications. Yet extant library classifications remain dominant within the library world for the simple reason that a bureaucracy exists that provides the Library of Congress Classification (LCC) and Dewey Decimal Classification (DDC) subject headings and call numbers for almost any book a library in the English-speaking world might acquire.

It is entirely possible, though, to develop a subject classification system that is much easier to apply than existing library classifications. The key to such a classification has long been appreciated within the academic field of knowledge organization. Rather than enumerate complex subject headings as LCC and DDC do, such a classification would allow the classifier to “build up” a subject heading by combining simpler terminology. Such a system can also be designed from the start to serve museum artifacts—but it can also be flexible enough to treat books, archival documents, and works of art. Such an approach is generally termed “synthetic.”

Vanishingly few museum curators have training in information science, or information scientists in museum curation. Curators understandably shy away from either mastering existing subject classifications or from hiring experts who know little about museums. But a museum curator can readily appreciate that they are holding, say, a golden axe used as a symbol of kingship. If they can build a subject classification combining terms such as “golden,” “axe,” “symbolize,” and “king,” subject classification will be straightforward. It will certainly be much easier than if the curator had to search for the complex subject that most accurately captures the essence of a particular object.

2. Motivation

Zoller and DeMarsh (2013) note that museums have traditionally focused on documentation—providing minimal information on what the museum possesses and where it came from—rather than cataloguing, which provides users with multiple access points. They cite a 1984 report of the American Alliance of Museums that urged, “Information sharing among museums … similar to the library information system that can locate all the books anywhere in the country on a certain subject is an ultimate goal” (American Alliance of Museums, 1984). The authors perform a survey and find a widespread belief that cataloguing requires curatorial expertise in a field. The authors worry that curators lack time and that experts in cataloguing may not wish to offend curators. They further note that cataloguing gets little attention in museum training. Only a small minority of museums had a position for cataloguer, and it was rare to want or require LIS training. One oft-voiced concern was that cataloguing would not respect the uniqueness of a collection. An approach to cataloguing that was easy to master and yet allowed uniqueness to be captured would clearly be advantageous.

Museums around the world have launched websites. But these are only rarely useful in guiding users, and especially professional researchers, to appreciate the detailed contents of a museum’s collection (Menard, Mas, & Alberts, 2010). Indeed, these websites often do not strive to do so. Menard, Mas, and Alberts worry that when information is provided on artifacts, this is often jargon-ridden and obscure. This suggests the value of employing basic concepts: concepts that have broadly shared understandings across different groups of experts as well as non-experts. These authors note that some sort of synthetic classification could allow museums to provide much better information regarding their possessions. They further note that few tools are available to museums wishing to pursue such an approach.

Museums were in the past often organized chronologically, and their artifacts thus classified primarily by time period. Museums (like art galleries) are increasingly organized thematically. It is thus of increased importance that their artifacts be classified by subject. This will among other things increase the ability of curators to identify pieces to borrow for thematic exhibits.

3. Basic concepts

There has been a huge debate within Knowledge Organization as to whether it is possible to identify controlled vocabulary that can be understood in similar ways across disciplines or cultural groups. This debate has obvious implications for the museum community: is it possible to provide descriptors for similar artifacts from quite different societies such that all potential users could understand what these descriptors mean? The solution I have proposed focuses on “basic concepts” (Szostak, 2011). Whereas complex terms such as “globalization” may be understood in diverse ways, basic concepts such as “trade flows” or “American movies” are subject to broadly similar understandings across individuals or groups. And thus a synthetic approach to subject classification in which basic concepts are combined can potentially allow users from different disciplines or cultures to find what they are looking for in museums and libraries. “Adobe” is a familiar term, but many will not know that it is (clay)(for)(building), whereas “clay” and “building” are broadly understood concepts.

Museums hold some artifacts because they are typical—say, the sort of sword used in a particular army—and other artifacts because they are unusual—say, a golden sword used for ceremonial purposes. Note that a synthetic subject string employing basic concepts can serve to identify both. A user seeking to compare a particular kind of artifact across time or place can thus readily identify these artifacts in many museum holdings. But a user interested in unique items can enter a very precise search string. We will provide dozens of examples below of possible subject classifications of museum artifacts.

Synthetic classification will prove especially useful for artifacts whose title may only be fully appreciated by experts in a field. We would want anyone searching for “tools for scraping” to find the “beamer,” a bone implement used for scraping by indigenous peoples in the Americas. Another example is Abrader: (tool)(for)(smoothing). To the unfamiliar, Apache Tear will hardly signify (round)(nodules)(obsidian). Synthetic classification is also particularly useful when terminology is ambiguous. Someone searching for awls used in working leather would like to readily distinguish these from awls used for working wood.

The “armor slat” will likely be appreciated only by experts in Native American warfare. It is a flat wooden piece, tied to others, employed as armor. Similar armor may well have been used in other parts of the world, but be referred to by different terminology. It is thus useful to render this compound as (flat)(wood)(for)(armor), where armor is itself coded as (protection)(for)(war). Note that there were other kinds of Native American armor. Some used wooden rods tied together; this similarity with slat armor will be captured by a synthetic approach. Many used hides, sometimes hardened with other substances, sometimes padded with cotton; again, a synthetic approach captures both similarity and difference.

Museum artifacts might be usefully classified in terms of their purpose, material composition, and (for some items) methods of manufacture (provenance is beyond the scope of this paper, though it should be noted that a synthetic approach is also useful in designating the group that created an artifact or the place of creation: say, along a lake in southeastern Europe). A synthetic approach is useful for all three: (axe)(for)(war); (wooden)(shaft)(steel)(head); (mass)(produced). A synthetic approach also allows these to be combined into one longer subject entry. This approach would allow the classifier to stress the most important characteristics of a particular artifact—but necessarily then encouraging one or two of the three elements to be ignored. Such an approach may also better identify artifacts whose specialness lies in unusual combinations of the three elements: (golden)(axe).

4. Ease of use

It should be stressed that a synthetic approach employing basic concepts is potentially easy for both curatorial staff to implement and for museum users to navigate. Both can have recourse to combinations of basic concepts. Both could be provided with a universal thesaurus that would guide them toward controlled vocabulary in cases where there are close synonyms.

Curators, with their detailed understanding of artifacts, should be able to classify—or supervise classifiers—without needing detailed training in information science. The inherent flexibility of the approach allows each museum to signal the unique attributes of each object. Yet a simple but shared vocabulary nevertheless facilitates searches across institutions.

5. The necessary controlled vocabulary

Such an approach to subject classification requires that the museum curatorial staff have access to controlled vocabulary. Many sources can be combined to this effect:

• Various sources, such as the Art and Architecture Thesaurus (Getty Research Institute, 2014) and Nomenclature (Bourcier & Dunn, 2015) already provide hierarchically organized lists of the sorts of objects museums are most likely to seek to classify.

• These lists, primarily of things, can be supplemented by classifications of verbs and of adjectival/adverbial qualifiers. In my own classificatory efforts, I have found that a list of some one hundred basic verbs can in combination with each other or with things or adverbs generate the thousands of verbs employed in the English language. I have also identified a few dozen key adjectives and adverbs. These can be of particular value in clarifying the shapes and sizes (morphology) of museum artifacts. See my Basic Concepts Classification (Szostak, 2013).

• The AAT or Nomenclature naturally focus on the artifacts themselves. But a subject classification will wish to grapple with the purpose or meaning of objects. And for this we need a classification of the subjects pursued in social science and the humanities. Here too it proves quite straightforward to capture key terminology within quite compact hierarchies. This is done in the Basic Concepts Classification (Szostak, 2013).

To be sure, much work remains to be done. And this is best done as part of a concerted effort to classify some extensive collection of artifacts. But the building blocks are there: both a logical structure and the basics of each of the necessary schedules of controlled vocabulary terminology.

6. Some examples of synthetic classification of museum artifacts

Museums hold a huge diversity of items. One way to sample this diversity is to consult the “highlights” placed online by a couple of the world’s leading museums. The choice of both highlights and famous museums biases our sample toward particularly remarkable artifacts. Since these are likely more difficult on average to classify than more mundane holdings, this bias may in fact be desirable in establishing the feasibility of the synthetic approach. Nevertheless, it will be useful in future to apply a synthetic approach grounded in basic concepts to other museum collections.

Museums contain some objects valued for their typicality and others valued for their uniqueness. A synthetic approach grounded in basic concepts is useful for both types of artifact. For the former, it allows us to capture the precise nature of an artifact: its design, materials, and use. For the latter, it allows us to also capture unusual design or decoration or purpose or association with people or events.

The Smithsonian Institution in Washington, DC, highlights 3,300 artifacts on its website (Smithsonian, 2014). Note that the website also provides access to media, lessons, primary sources (which seems to be a general category that includes artifacts, webpages, and so on), websites, reference materials, and worksites. There is thus presumably much value even within the institution in having a common classification for items that might be found in museum, archive, library, or online.

We look closely at the first ten artifacts (as of March 1, 2014). Our focus here is on how a concept string could capture important elements of the subject. We would expect that each item would also be classified by date and title and (if possible) creator. We shall find that date and title would often be critical to a subject search, and would thus advocate that these elements be available for subject searching.

The first item is a “1778–1943 Americans Will Always Fight for Liberty” Poster. The chain (Poster)(encourage)(fight)(war) captures its key elements. Users would find the poster if interested in encouragements to fight or posters about war. The second item, “Hamons Court” Neon Sign, can be captured with (neon)(sign)(motel), and thus accessed by those interested in both neon and motels, but especially their combination. The third item is the well-known photograph “Migrant Mother” by Dorothea Lange. It might be captured by (photograph)(migrant)(worker)(mother)(poor)(1930s), but this photograph is well known because it captures the woman’s desolation, presumably regarding her limited ability to provide for her family. So we should add at least (despair) and perhaps best (despair)(because)(poor) to the chain [replacing just “poor”]. The fourth item is “Towne of Boston” Flintlock Musket, or (flintlock)(musket) where musket is recognized as a type of gun [note that the name and date would also be captured, and could be searched for]. The fifth, “What Hath God Wrought” Telegraph Message is (telegraph)(message)(transcribed). [Again, much of its essence is captured by date and title.]

The sixth, “Woodsy Owl” Anti-Pollution Poster, can be captured by (poster)(opposing)(pollution). The seventh, 16-Pound Bar Shot, is (ammunition)(against)(ships). The material should perhaps also be coded. The odd shape, with enlarged heads on both ends of a narrow rod, deserves mention but is admittedly hard to capture: (enlarged)(heads)(on)(narrow)(rod). The eighth, 1848 gold nugget, merits special consideration as: (first)(gold)(nugget)(found)(California) [1848 would be part of the classification, and should be searchable in a subject query.] The ninth, 1874 Remington Type Writer, is (first)(Remington)(model) (typewriter). The tenth, 1876 Ellen Harding Baker’s “Solar System” Quilt, can be rendered (quilt)(solar system)(made from)(wool)(cotton)(embroidered)(silk). Note that the material of manufacture is of critical importance here.

The simple point to stress is that for each of these ten quite different artifacts, their subject is best captured by two or more distinct elements in combination. Combining only basic concepts still allows us to capture the uniqueness of each in a string of only several elements in length.

The British Museum in London (2014) does not have a highlight list online but rather encourages search by cultures (or along other dimensions). Given the relative modernity of the items highlighted by the Smithsonian, it makes sense to start with the first (alphabetically) of ancient civilizations. The first item under Achaemenid is chalcedony cylinder seal. Each term here is a basic concept (though admittedly chalcedony, a compound crystal of quartz and moganite, requires a very detailed classification of crystals, or must receive a compound rendering itself). It could be added that such a seal was only used by officials, or that it is highly detailed and high quality. “The most common subject is a crowned figure wearing Persian dress, here shown fighting a lion alongside a hero in Babylonian dress shown fighting a bull.” We could capture these key elements also: (crowned)(male)(fighting)(lion)(beside)(Babylonian)(male)(fighting)(bull).

There are several Achaemenid artifacts that lend themselves to a short compound classification: deep silver bowl; silver dish; agate cylinder seal; blue chalcedony cylinder seal; calcite jar, and so on. In each case, we might choose to specify the type of decoration or manufacturing process if these were deemed noteworthy. A set of four sequential silver bowls is described respectively as decorated with petals (two of them), a lotus flower, and embossed gold figures. The second of the four required hammering, embossing, and engraving. It is this combination that gives the bowl its special nature.

There are several small figurines or statuettes. A classification system will have to clarify the difference, if any, between such terms: Wikipedia, it might be noted, treats figurines as a kind of statuette and statuettes as a kind of figurine. They can each be distinguished from larger statues by size: (small)(statue). They can apparently be distinguished from “ornaments” (another common descriptor for the British Museum) by a focus on animal, human, or deity. It would seem more important to distinguish these three types of statuette (or figurine) than to identify the distinction between statuette and figurine. It may thus be best to take a synthetic approach here, linking say a human statuette to the chain (small)(sculpture)(human).

The first “figurine” is a “bronze figurine of horse and rider.” This is also an easy compound. Special elements that might be captured include the artifact’s rarity and the facts that the rider carries a sword and wears trousers.

Many of the Achaemenid artifacts are from the Oxus treasure, a (likely) temple excavated near the Oxus River. This deserves to be noted in the subject string; “Oxus treasure” can in turn be translated into its own concept string.

Several Achaemenid artifacts involve fragments of walls or columns. We certainly need a common vocabulary for the elements of a column, and perhaps parts of walls and ceiling (think of the importance of the spandrel—the parts that link a square room to a round dome (or an arch to a wall)—in the history of art and architecture). It may be less useful to classify the type of fragment (since searches by type of fragment may be unlikely, except perhaps by a curator interested in how to display a fragment). One noteworthy artifact of this type is “Glazed brick relief panel.” We would wish to indicate that it is part of a larger frieze depicting eighteen immortal bodyguards; the fragment depicts one such figure.

Another notable artifact is “Fluted silver drinking horn with griffin.” There are two holes in the griffin’s chest that could be closed by fingers for drinking or opened for pouring. The concept chain (holes)(for)(drinking)(or)(pouring) could identify the piece for anyone looking for that precise feature (even if they used a slightly different combination).

There are several “plaques.” These are perhaps best identified as (flat)(sculptures), though “flat” may miscommunicate the fact that plaques have relief.

There is a “Stone relief from the audience hall at Persepolis.” Only in the description do we find out that it depicts “Susian” guards, so-called because they are similar to glazed figures found in Susa, and thought to be immortal guards: this important element should be captured, perhaps as: (guards)(likely)(immortal)(similar form)(Susa).

All other Achaemenid artifacts lend themselves to fairly straightforward combinations of basic concepts (see Appendix).

What of a quite different time period? The British Museum has 161 highlights from Medieval Europe. These are often quite similar in type to the earlier artifacts: drinking vessels, figurines, and so on. It does become apparent, though, in surveying the Medieval collection that careful distinctions need to be drawn between terms such as chalice, beaker, and cup.

More detail tends to be known regarding each medieval artifact, and thus the potential subject headings are much longer. For example, the Clephane horn is carved elephant tusk, is probably from southern Italy, and was used in Carslogie Castle Scotland. Thus (horn)(elephant)(tusk)(probably)(carved)(south)(Italy)(used)(Carslogie Castle), where “horn” would be defined as (instrument)(for)(sending)(loud)(signal), and Carslogie Castle would be linked to Scotland. [Harpring (2002) urges us to use a word like “probably” in situations where there is controversy regarding the subject of a work. This is easily done synthetically.] The subject chain would need to be even longer to capture the pictures carved into the horn in three sections: a horse race at the Hippodrome, staged hunting scenes, and an official ready to give a prize to the victor.

One interesting artifact is the Gemellion: a paired wash basin so that water could flow through holes from one to the other. (paired)(wash basins)(with)(holes) might suffice as a subject heading. A user with an idea that such a device existed should be able to find it.

The “Chatelaine plate” also suggests a complex subject chain: (to be)(hung)(from)(belt); (eagles)(eating)(fish); (tinned)(to appear)(silver). The latchet dress fastener would require a lengthy chain to describe its peculiar mechanism, but perhaps (dress)(fastener)(unusual)(design) would suffice.

There are numerous coins in the collection. These could each be classified in terms of material, place of issue, and figures on the surfaces.

There are many crosses and other religious artifacts. These might all be designated (associated)(Christianity) and where possible their use in particular ceremonies indicated. The “Pilgrim sign of St Thomas Becket” could be denoted as (souvenir)(pilgrimage)(Canterbury)(denoting)(Beckett)(body)(under)(shrine). The likelihood that this souvenir may denote the movement of the shrine can likely be left to the description.

The Seal-die of Isabella of Hainault is rare, as seal dies were generally destroyed at death: (rare)(seal)(die)(queen)(with)(scepter)(and)(fleur de lis)(quality).

As noted above the main challenge in classifying medieval objects is that we know many interesting things about many of these. These would ideally all be classified, though there may be administrative challenges in doing so. These challenges may be lessened by the acceptance of a common controlled vocabulary that is easy to master and apply.

A thesaurus

The advantage of controlled vocabulary is that users can know what terms the classifier has employed. The disadvantage is that the user must first identify these terms. If controlled vocabulary terms are organized into small and logical hierarchies, then it is easy to identify desired terminology. This process can be further aided by a visual user interface that allows the user to quickly move through organized hierarchies in order to identify desired search terms. This is difficult within existing library classifications because hierarchical organization is obscure at best. But the proposed system lends itself to hierarchies that are both logical and compact.

Nevertheless, it would be useful to provide some sort of thesaural interface such that a user inputting non-controlled vocabulary can be guided to the correct search terminology.

Pre-coordinated and post-coordinated

Another debate within Knowledge Organization involves the relative advantages of “pre-coordinated” subject classifications like LCC and DDC in which classifiers must choose from a preexisting set of subject headings, and “post-coordinated” systems such as the one proposed in which classifiers can synthesize simpler terms. The latter, it is appreciated, can have much shorter and logically organized schedules of controlled vocabulary. But it is feared that there is a lack of precision: users seeking works on “philosophy of history” may be guided to works on “history of philosophy.” This weakness is not inherent in post-coordination but depends on what sorts of search algorithm are employed (Szostak, 2015). I worked with a group of computer science students in 2015 who developed a set of search algorithms that users could employ depending on how important the order of search terms was to them.

7. The Semantic Web

The idea behind the Semantic Web is that information in websites can be coded in a manner that allows computers to identify synergies across databases. If one website asserts that swans are birds, and another that birds have wings, the computer can conclude that swans have wings. For present purposes, the first key point is the coding: the developers of the Semantic Web appreciate the limitations of keyword searching—if the different databases in the example above used different terminology for “bird,” no deduction would be possible—and thus advocate coding with the use of controlled vocabulary. The second point is that the form of this coding centers on “RDF triples” of the form (subject)(predicate or property)(object). That is, databases are to be coded in terms of combinations of things, verbs, and adverbs/adjectives. This is exactly the sort of approach recommended above for “coding” the subjects of museum artifacts—though admittedly our subject strings will often be longer than three terms in length. It is possible, though, to translate longer strings into combinations of RDF triples. We can thus potentially provide subject classification of museum artifacts that facilitates not only human users but searching by computers. And a controlled vocabulary adopted in the museum community would likely be employed more widely, for the Semantic Web community has signally failed to achieve consensus on controlled vocabulary (Hart & Dolbear, 2012).

8. Appendix

An Appendix to this paper provides discussion of many more Achaemenid artifacts. It also discusses how a synthetic approach would address dozens of items on the lists of descriptors employed by the U.S. National Park Service; this approach provides useful clarifications of terms that are either ambiguous or obscure or culturally specific. The Appendix also applies a synthetic approach to some items from the U.S. National Gallery of Art, in order to show that the approach is useful for works of art as well. And the Appendix also links to a translation exercise in which the elements of Iconclass, a subject classification of works of art, were translated into the terminology of the Basic Concepts Classification, again with a considerable increase in clarity.

The committee on documentation of the International Council of Museums (CIDOC) has produced the CIDOC Reference Model (2013). As a guide to metadata elements, it provides as much detail on museum practices such as acquisition and attribution as on subject description. Since there is value in utilizing the same vocabulary across metadata elements, I also show in the Appendix how the various elements of the CIDOC model can be translated into the synthetic terminology of the Basic Concepts Classification.

The Appendix is available at: https://sites.google.com/a/ualberta.ca/rick-szostak/publications/synthetic-classification-of-museum-artifacts-using-basic-concepts

9. Conclusion

The purpose of this paper was to provide theoretical and empirical support for the conjecture that museum artifacts (and indeed items across the GLAM sector) can best be classified utilizing a synthetic approach grounded in basic concepts. It is likely also well suited to the Semantic Web and could thus support machine searching across GLAM and beyond.

Perhaps most importantly, a synthetic classification was developed for a variety of works from an international selection of museums. These classification exercises hopefully establish the feasibility of providing very detailed classifications of museum (and gallery) items by compounding basic terms in concept chains of manageable length. Museums could decide how detailed they wished to be in their classification. Notably, the basic concepts employed came not from a classification designed just for museums but from a general scheme. This allows users to search for items related to any human or natural activity or thing with which they are interested.

References

American Association of Museums. (1984). Caring for Collections: Strategies for Conservation, Maintenance and Documentation. Washington, DC.

Bourcier, P., & H. Dunn (and the Nomenclature Task Force). (2015). Nomenclature 4.0 for Museum Cataloging: Robert G. Chenhall’s System for Classifying Man-Made Objects. American Association for State and Local History.

British Museum. (2014). Explore/ Highlights (By culture). London. Consulted September 7, 2015. Available https://www.britishmuseum.org/explore/highlights/cultures_index.aspx

CIDOC (Conceptual Reference Model, Definition of the). (2013). Produced by the ICOM/CIDOC Documentation Standards Group, continued by the CIDOC CRM Special Interest Group, Version 5.1.2, October 2013.

Getty Research Institute. (2014). Art and Architecture Thesaurus. Consulted September 7, 2015. Available http://www.getty.edu/research/tools/vocabularies/aat/

Harpring, P. (2002). “The Language of Images: Enhancing access to images by applying metadata schemes and structured vocabularies.” In M. Baca (ed.), Introduction to Art Image Access: Issues, Tools, Standards, Strategies. Los Angeles: Getty Research Institute, 20-39.

Hart, G., & C. Dolbear. (2013). Linked Data: A Geographic Perspective. CRC Press.

ICONCLASS (2014). Consulted September 7, 2015. Available www.iconclass.nl.

Menard, E., S. Mas, & I. Alberts. (2010). “Faceted classification for museum artefacts: A methodology to support web site development of large cultural organizations.” Aslib Proceedings 62(4/5): 523-532.

Smithsonian Institution. (2014). History Explorer. Washington, DC. Consulted September 7, 2015. Available http://historyexplorer.si.edu

Szostak, R. (2011). “Complex concepts into basic concepts.” Journal of the American Society for Information Science & Technology 62:11, 2247-65.

Szostak, R. (2013). Basic Concepts Classification. Available at: https://sites.google.com/a/ualberta.ca/rick-szostak/research/basic-concepts-classification-web-version-2013

Szostak, R. (2015). “A pluralistic approach to the philosophy of classification.” Library Trends 63:3, 591-614.

Zoller, G., & K. DeMarsh. (2013). “Museum Cataloging from a Library and Information Science Perspective.” Art Documentation 32:1.