Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

Chapter 3: Describing Instances Chapter Author: Kimra McPherson ([email protected])

Contents  3.1. Describing Instances: An Overview ................................................................................................... 1  3.1.1. Levels and Contexts of Description ........................................................................................................... 2  3.1.2. Tradeoffs of Description ................................................................................................................................ 5  3.1.3 Some Properties of Description ................................................................................................................... 6  3.2. Process of Description ............................................................................................................................ 8  3.2.1. Identify and scope the thing to be described ........................................................................................ 8  3.2.2. Study it to identify its important properties or features .................................................................. 8  3.3.3. Compare it with other things like it and unlike it ................................................................................ 9  3.3.4. Select or develop a vocabulary for using “good” terms..................................................................... 9  3.3.5. Create the descriptions either “by hand” or with computational help ..................................... 10  3.3. Describing Multimedia and Multi­Dimensional Instances ..................................................... 10  3.3.1. The Challenges of Multimedia Instances ............................................................................................... 10  3.3.2. History of Describing Multimedia Instances ....................................................................................... 11  3.3.3. Modern Methods of Describing Multimedia Instances .................................................................... 11  3.4. Who Chooses Descriptions? ............................................................................................................... 14  3.4.1. Describing is Challenging and Contentious .......................................................................................... 14  3.4.2. People Choose Descriptions ....................................................................................................................... 14  3.4.3. Information Professionals Choose Descriptions ................................................................................ 15  3.4.4. Businesses Choose Descriptions ............................................................................................................... 15  3.4.5. Governments and Interest Groups Choose Descriptions ............................................................... 16  3.4.6. Computers and Other Automated Processes Choose Descriptions ............................................ 16  3.5. What Makes a Good Description of an Instance? ....................................................................... 17  3.5.1. No Descriptive Language is Perfect for Every Situation ................................................................. 17  3.5.2. Traditional Principles .................................................................................................................................... 18  3.5.3. Business and Standards‐Related Principles ........................................................................................ 18  3.5.4. Computational Principles ............................................................................................................................ 19 

 

3.1. Describing Instances: An Overview What are “cylinder one-ers,” “coke bottles,” “golder wipers,” and “round one-bricks”? Turns out, they can all be the same thing—if you’re a 7-year-old playing with Legos. One afternoon, writer Giles Turnbull asked his young son and a group of his friends to share their descriptive names for Lego pieces. Each of the four boys turned out to have his own set of personal descriptions for the tiny building blocks. Some descriptions were based on color alone (“redder”), some on color and shape (“blue tunnel”), some on role (“connector”), some on

Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

common cultural touchstones (“light saber”). Others, like “jail snail” and “slug,” seem unidentifiable—unless, of course, you happen to be inside the mind of a particular 7-year-old kid. But that typically doesn’t matter, as long as the descriptions are full and clear enough to allow the little builders to understand each other while they’re constructing miniature spaceships in the living room. “Within families, the communication is clear, and a clippy bit is most clearly a clippy bit,” Turnbull wrote. “Lego, like language, benefits from context” (Turnbull, 2009, para. 17). The Lego story is cute—who doesn’t love a good “kids say the darndest things” moment?—but it also illuminates many of the key issues in describing items for organization and retrieval. Coming up with descriptions is a natural and common act; it’s one of the major ways we make sense of the world and let each other know what we’re talking about. But how can we be sure we’re talking about the same things? The problem of creating good descriptions is something that’s been fundamental to the information disciplines for generations, with professionals ranging from librarians to web developers seeking methods of description that will let us organize the world, retrieve the things we want to find, and communicate effectively with others. Sometimes, those efforts have taken the form of bibliographic principles, controlled vocabularies, and formalized objectives. Other times, they’ve celebrated the messiness of user-generated, descriptive tags piled on by the online masses. In this chapter, we’ll examine some of the major issues around describing instances of various sorts, from a book to a song to a service system to an information need. We’ll look at the trouble spots where description becomes dicey, and we’ll present some principles for deciding when a given description is the right one for your needs.

3.1.1. Levels and Contexts of Description To start, let’s think about how we might describe something we’ve all seen and used in our lives: a chair. In your house, you might describe your chair by the room it’s in, e.g., “the kitchen chair.” But then you take it to a potluck dinner at a friend’s house, where all the chairs end up in the kitchen. So now it’s “my chair,” or “the wooden chair,” or “the folding chair,” or “the black chair with white trim.” Maybe you inherited it, so when you talk to your family, you call it “grandma’s chair.” Maybe you’ll get famous and the chair will be donated to a museum, where it’ll be listed as “[your name here]’s chair.” Or maybe you’ll sell it on eBay, where it’ll be “all-oak wooden kitchen chair mint condition—free shipping!!!!!!!!” It’s always the same thing, of course, a single chair. But as with any thing, either ordinary household object or complex information entity, to describe it precisely so that others understand what “it” is, you need to take different approaches depending on the context, the size of the collection of objects or items, and the audience for whom you’re describing. Description depends on your goal First of all, why are you describing something in the first place? Different activities or goals will require very different approaches to description, some more rigorous, some less so. Are you trying to organize a collection of your own things, for your own purposes? If so, you might pick out attributes and terms that mean something to you, that way, when you want to find things again, you’ll remember how you described them in the first place. For example, if you have a

Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

favorite chair, and only you care about which one it is, then describing it as “my favorite chair” tells you everything you need to know. But if you’re trying to organize information for others, you may need to be more deliberate in your approach. If you’re a librarian, for example, you’ll probably consider bibliographic objectives that will help patrons of your service identify, locate, and obtain the material they seek, perhaps following guidelines of the user tasks set forth by the Functional Requirements for Bibliographic Records: find, identify, select, obtain (Tillett, 2003, p. 5). If you’re a company, you might want to know how the other businesses with which you interact describe the same types of entities and interactions so that your information will mesh better down the line. Descriptions may also need to be read by computers or other machines, using a special kind of language that’s meant for machines and not humans. These descriptions may map to more natural human-language descriptions, and they require the same sort of goal-oriented creation in order to make sure the system performs as intended, without unpleasant surprises or confusion of data. Description depends on context. Let’s stick with the chair example. In our own home, the way we would describe a chair differs from the way we would describe that exact same chair at a friend’s house. Where we are, what our chairs look like, how much we know about the other chairs — all of these things influence the words we use to distinguish our chair from the rest. In the Lego story, context is key: The boys playing with the blocks were communicating with each other, in person, with the Legos in front of them. If they hadn’t been able to see the blocks the others were talking about, or if they had to describe their toys to someone who had never played with Legos before, their descriptions might have been quite different.

Description depends on the size of your collection. Frankly, if we only had one thing to describe, we could use whatever words we wanted to describe it. We wouldn’t need to distinguish it from anything else. If you only had one chair, there wouldn’t be any point in constructing a rigorous description for it; it’s the only chair you have. But things get interesting when the collection grows, often requiring more and more characteristics to tell one thing apart from the rest. Description, therefore, is highly dependent on scale—the size of your collection of things as well as the size of your intended audience of users. We’ll address audience in the next section. Consider the scenario where you take your chair to a friend’s house. Suddenly, the words you use to describe your chair start to depend on the properties of the other chairs in the room, too. If all the chairs are wooden, calling yours “the wooden chair” doesn’t help. Furthermore, if they’re all about the same color, “the brown, wooden chair” won’t get you very far either. So we have to start thinking about what other properties might distinguish our chair from the rest. “The brown, wooden folding chair” might do it; so might “the brown, wooden folding chair with a paisley cushion attached.” It all depends on how many chairs you have and what qualities set each one apart from the others. On a (literally) universal level, consider the case of poor Pluto, once our tiniest planet but since demoted to a mere “dwarf.” Back when astronomers thought there were only nine large objects orbiting our sun, it was easy to just call them all “planets.” But once it became clear that

Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

there were more than nine objects in this solar system and also that stars other than the sun had solar systems of their own, astronomers needed to develop more scientific and formal criteria for what was and was not an planet instead of defining the concept of planet by listing all the things that were called planets. Under this new definition of planet, Pluto no longer made the cut because it didn’t satisfy the requirement that a planet must “clear the neighborhood around its  orbit” (Foust, 2006). Creating descriptions that can keep pace with the growth of a collection has been an issue for librarians for years, as libraries moved away from describing simply “whatever came across a cataloger’s desk” (Svenonius, 2000, p. 31) to cataloging resources for a national and even international audience. But librarians aren’t the only ones dealing with the problem. Scientists who discover new life forms or celestial bodies, businesses trying to clarify the differences between 25 different kinds of printers, and amateur photographers who return from vacation with 600 digital photos and have to find a way to remember the difference between IMG0477 and IMG0478 all wrestle with the challenges of creating scalable descriptions. The point is this: The more items we have around, the more descriptors we likely need to distinguish our particular item from the crowd—and the greater the chance that we may need to impose some formal or controlled vocabulary on our descriptive language to make sure the terms we use always mean the same thing. This doesn’t just apply to concrete objects such as chairs; it also matters for things far more abstract, such as services (“the bank,” “the online-only bank,” “the online-only bank that lets me take money out of ATMs overseas without a fee”). As we saw in section 2.3.4, it’s not always easy to know how precise we need to be. When our collections grow, our language often must become more rigorous, moving from a simple glossary that sets definitions for terms in a particular domain to an ontology that not only defines terms but that also explicates the relationships between them. Description depends on your audience. The amount of knowledge your audience has about the thing you’re describing and the amount of knowledge that they share with the “describer” can affect the amount and type of descriptors you apply. Your friends might not know which chair in your apartment was “grandma’s chair,” but your family could pick it out instantly. Your friends might know that you always call the gray Lego a “silver plate,” but you might try a different description for someone who’s never heard you discuss Legos before. Consider what happened with the naming of the H1N1 virus — most commonly known as “swine flu.” For the medical industry, naming this strain of flu after the animal in which it was originally found was common practice. But the pork industry feared that its falling sales indicated that the general public took it to mean something else: that eating pig products could transmit the virus (Martin and Krauss, 2009). Some groups, meanwhile, might intentionally adapt different descriptions for the same thing to different audiences. A restaurant trying to sell its patrons on a nightly fish special might create a mouthwatering description of a Chilean sea bass dish, for example, while a sustainable seafood organization urging people to steer clear of the species might refer to it by its lessdelicious name, the Patagonian Toothfish. Similarly, what were once marketed as prunes are increasingly being labeled as dried plums to attract a younger audience of consumers looking for a healthy snack food rather than as a “regularity” aid for constipated old folks.

Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

Audiences also change in size, and as they scale, the efficacy of your original descriptions may change. For example, a business may start as a community mom-and-pop shop, where regional or community-specific descriptions for things suffice. But what happens if that company begins to grow, reaching out for customers in other regions, and, eventually, companies? The descriptions that worked just fine when the business was limited to a single community may need to be redefined to make sense to a wider audience.

3.1.2. Tradeoffs of Description With all of these contingencies, one thing is clear: Describing things is never easy, even given the best of circumstances. If you need your descriptions to make sense to a wide variety of people in a range of different circumstances, or if you anticipate that a collection will grow and want to begin with scalable descriptions, it can be even tougher. Thus, there are a number of tradeoffs related to designing description, and we discuss a few of these below. Simplicity vs. Complexity There are times when a simple description might be enough: You want to talk about the wooden chair in a sea of plastic ones or the red Lego in a pile of blue pieces. But as we’ve seen, descriptions can get very complicated very quickly. Simple descriptions have their benefits: They’re quick, they’re relatively easy for anyone to create, and often, they can communicate enough information to get your point across. However, creating a more complex description can have benefits down the line, both for communicating information more precisely and for allowing other people to understand it independent of your explanation. Organization Time vs. Retrieval Time With all the considerations that go into creating a good description, one might ask: Why should the organizer of the information have to do so much work just to be understood? Shouldn’t the person seeking the information bear some of the burden? Turning to the chairs again, here are two scenarios that illustrate this tradeoff: Scenario One: You determine that the chair you want is the only wooden folding chair with a cushion. Then, you tell your friend to go get you the wooden folding chair with the cushion. You have to think through which words will let your friend know which chair you want, but your friend returns the chair with no trouble. Scenario Two: You call your friend on his mobile phone and ask him to bring you a chair. Your friend asks, “Which one?” You say, “the wooden one.” That narrows it down a little, but not enough. So your friend says again, “Which one?” “The folding one,” you reply. Still, your friend’s left with a choice between two chairs and again asks, “Which one?” “The one with the cushion,” you say. Eventually, your friend brings you the chair. The difference between these two scenarios is the essential difference between devoting time to creating a good description up front, at the point of organization, and putting the work in later, at

Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

the point of retrieval. Both get the job done; in each case, you get your chair. But the point at which the work is done is different, and the second scenario really only works when you have several chances to communicate with the person retrieving the chair for you. If you were leaving a request in writing, you’d have no choice but to create a more precise description off the bat or risk getting stuck with someone else’s chair. Automated Description vs. Human Description In section 2.5.2, we introduced the concept of the semantic gap—the space between the meaning we can assign to some description and the functional but not “meaningful” descriptions computers and other automated processes can assign. Both types of description have their place, and it’s important to consider which one you want to use for a given situation. Human description can potentially convey a greater level of meaning; it can also be expensive and time-consuming to create, and depending on the skill level of the person who’s creating the description, it can be inconsistent with the controlled language or descriptive principles you wanted to follow. Automated description, meanwhile, has one major benefit right in the name: It’s automatic (or at least mostly automatic), so it can be assigned without significant human effort. The downside, however, is that it’s much more difficult to imbue with meaning. The digital signature of a song or the date on a digital document can be useful descriptors, but they don’t inherently convey much about what something is “about.”

3.1.3 Some Properties of Description In the rest of this chapter, we’ll turn our attention to the ways different things have been described, both traditionally, in libraries and collections, and more recently, on the internet and in other electronic contexts. No matter what you’re describing or who’s creating the description a few properties should be considered. Name As we discussed in the previous chapter, naming something is often the gateway to describing it more fully. But it’s also challenging and contentious: names can mean different things to different people, and choosing the right one is a job unto itself. Key decisions include whether the name will be controlled or natural-language; whether there will be any way to “map” naturallanguage terms to a common name later; whose version of the name will be official or preferred.; and who will be responsible for making and enforcing those choices. Physical properties Describing perceptible physical properties can start to distinguish a number of like objects from each other. If a particular Lego is blue, it’s set apart from all the not-blue Legos; a square Lego is physically different from a round one. (Of course, if someone is color-blind, using color as a descriptor is not as effective.) For all their benefits, however, physical properties can be merely a start to describing something. In many cases, physical properties stick to the surface; they often reveal little about what something is, what it means, or when and why it was created, and for information components that don’t take an actual, tangible form, physical properties can be entirely separate

Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

from their information and can change without fundamentally altering the information being offered. For example, if this paragraph were highlighted in bold or cast in 24-point font, it would still say exactly the same thing. Sometimes, though, physical properties can lead to sufficiently rich descriptions. Consider the case of onomatopoeia, or words that sound like the thing they’re trying to describe — “quack” for the sound a duck makes, or “fizz” for what happens when you open a soda bottle too soon after shaking it. Marvel Comics is known for trademarking “thwip!”, which is the noise Spiderman’s web makes when deployed. And presentation occasionally sends us clues about information: Most of us recognize bold text as something to which extra attention should be paid, and a greater-than-normal number of lowercase letters, ragged lines, and centered text could be a clue that a particular piece of text is a poem. Still, most of the time, physical properties are just one dimension of description, one that varies greatly in its usefulness. For further meaning, we frequently turn to cultural, contextual, and structural properties. Cultural and contextual properties An object can have a number of different cultural or contextual properties, and they can be powerful descriptors. When the Lego boys all called one piece a “light saber,” for example, nobody had to tell them to think about Star Wars. “Light saber” was just the obvious word for a long, neon tube with a handle. When a description plays on a common touchstone, it can be instantly memorable. But culturally specific descriptions can also present interesting complications. Someone who’s never seen or heard of Star Wars, for example, would probably have a hard time understanding why some particular piece of plastic in the Lego set would be called a “light saber.” And in some cases, cultural descriptions can evolve on a trajectory of their own. For example, a particular type of geometrically patterned Turkish rug came to be known as a “Holbein carpet” after the German painter Hans Holbein, who often depicted the rugs in his work. But the rugs themselves existed (and were even painted by others) long before Holbein made his renderings. And a particular square, shaggy women’s haircut might always be called “the Rachel,” even though the TV show Friends—whose Rachel character spawned the popular ‘90s look—has been in syndication and off the primetime schedule for years. Contextual properties, too, can be tricky. Some qualities of an instance are noticeable in one context but disappear in another. Sarcasm, for example, is much easier to identify in speech than it is in written communication — a fact that led some enterprising folks to develop the SarcMark, a cheeky punctuation mark that claims to “ensure that no sarcastic message, comment or opinion is left behind.” Along with physical properties, some cultural and contextual properties of items — such as the name of the artist who painted a particular picture or the date a webpage was created — often end up stored as metadata. We’ll discuss metadata more in chapter 4. Structural properties The way something is assembled, the links between it and other things, and its internal structure can also help develop a unique description. For example, how one website is linked to another can add a level of precision or understanding to the descriptions of those sites. We’ll discuss these issues more in chapters 7 and 8.

Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

3.2. Process of Description Whatever it is that we’re describing, following a common set of steps can help us to know we’re creating a solid description. This process has evolved over time, and its application may vary slightly in different contexts, but in general, the key pieces of the process remain the same. In this section, we’ll discuss the various steps in the process, their origin in library science, and their applicability and expansion to an increasingly digital world.

3.2.1. Identify and scope the thing to be described What is this “thing,” this instance that we’re attempting to describe? If you read chapter 2, you know this is a harder question than it might at first appear. Do you want to describe a physical book, or the content within it? Are you describing an entire bank service, or just its online homepage? Identifying as precisely as possible the thing you want to describe is the first step to creating a solid description. For more on this, see section 2.1.1.

3.2.2. Study it to identify its important properties or features A first step in creating a successful description is deciding which of its properties are worth recording. Often these are its most essential properties, the ones that could best differentiate a given thing from a crowd. What features are considered important or essential, though, is a matter of judgment that requires careful consideration. Ever since the creation of the very first library catalog, different approaches have been taken to determining which properties are essential. As William Denton writes in his brief history of library cataloging, “There were many different kinds of catalogs, some very good and others just inventory lists. Any way you can think of to use books was probably used, including by size, color, or the name of the person who donated them” (Denton, 2007, p. 36). In library science, a few key properties tend to be described first. Author, title, and subject are three such common properties, each of which comes along with its own rules. But Denton notes that every kind of catalog has its drawbacks: In subject catalogs, it can be easy to find all books on, say, the French revolution or monarch butterflies, but discovering their authors can take more work. In an author catalog, on the other hand, all works by a given writer are grouped together, but determining the subject of each of those works is not always easy. Of course, these divided catalogs can be merged to create complex bibliographic systems that account for all properties and include surrogate records that enrich description for the purpose of finding things. What properties are important depends on your broader goals. If you’re curating a collection of a single author’s works — or, say, trying to organize your own academic papers on your computer’s hard drive — then a successful description likely prioritizes titles, dates, or other variations over the author’s name, which would be the same for every instance. In Document Engineering, Glushko and McGrath (2005) suggest the existence of a “component model” for describing items or information components in some domain. These components can then be assembled and reassembled in various orders or hierarchies depending on the perspective or priorities of a particular project, dubbed “document assembly models.” By changing the order or hierarchy of the descriptive components, different priorities or perspectives

Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

are revealed. For example, a publisher might want to look at the roster of all of its authors, where the authors’ names are described according to a particular format; an author, meanwhile, might want to keep track of all the books he or she has published through different publishing houses.

3.3.3. Compare it with other things like it and unlike it We know now that it can be hard to determine what something is. Sometimes, knowing what it isn’t can help define what it is and create an appropriate description. So we want to compare the thing we’re describing with other things. For example, are all of your Legos square but only one blue? Were all of your files created in December, except for the one you last saved in January? Knowing where something fits in the grander organizational scheme is one of the most important aspects of creating principled descriptions, and it can even be considered the charter of information organization as a whole. According to Svenonius, “The essential and defining objective of a system for organizing information, then, is to bring essentially like information together and to differentiate what is not exactly alike” (Svenonius, 2000, p. 11). When this kind of comparison starts to look more like sorting—all the square, blue Legos in one place—it’s called categorization. We’ll be discussing the ins and outs of categorization more in chapters 5 and 6. For now, though, just assume that there are some properties of any thing that make it a good fit for one category, some that make it fit another, and knowing which of these makes the object most easily identifiable is a key part of the process of creating a good description.

3.3.4. Select or develop a vocabulary for using “good” terms What words should you use to create your description? Once again, this depends on the context and purpose for which you are creating a description. If it’s only for you, you should use the terms that you’ll remember; more often, however, your descriptions will need to be understood by some outside audience, and so identifying a set of “good” terms to use in your description is essential. Essential, but not easy. “It would seem that the most colossal labor of all involved in organizing information is that of having to construct an unambiguous language of description,” writes Svenonius, “a language that imposes system and method on natural language and at the same time allows users to find what they want by names they know” (Svenonius, 2000, p. 14). We know from our discussions in chapter 2.5 that “natural” language doesn’t always get the job of identification and description done, purely because we use a wide variety of words for the same thing or concept. Instead, creators of good descriptions within principled organizing systems often turn to controlled vocabularies to create the set of standard, acceptable terms to be used in a description. These controlled vocabularies can define the precise meanings of the various words you use to describe things. In your own system, you may just need to know that “San Jose” refers to “San Jose, California” and not “San Jose, Costa Rica.” But if you’re describing things for others, you may want a more structured way of sorting through these words — often a glossary, dictionary, or ontology, which can provide various levels of structure and meaning.

Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

3.3.5. Create the descriptions either “by hand” or with computational help How this final step is done—how the key properties are recorded—has changed significantly over time, from hand-written cards kept in a catalog to online tagging fields that apply a description nearly instantaneously. Which method of recording and scoring descriptions you choose has much to do with the type of organizing system you’re using and its goals and intended audience. We’ll discuss some of the ways descriptions are encoded and stored in chapter 4, which looks at metadata. And in chapter 11, we’ll delve in to the ways different types of descriptions—whether created by hand or by machine—affect information retrieval.

3.3. Describing Multimedia and Multi-Dimensional Instances 3.3.1. The Challenges of Multimedia Instances Many traditional library science principles were developed for describing text and other physical media. But relatively quickly, those principles had to evolve to deal with different types of objects people wanted to catalog, from paintings to mp3s. The realm of describing multimedia instances has broadened greatly with the rise of the Internet and digital media: Now we have to figure out the appropriate descriptors for instances of many more types than ever before. The ease of creating more and more media also affects the scope of the problem. We can take thousands of digital pictures on a trip because we are far less limited by the size of our camera’s memory cards than we were by the cost of physical film, the number of exposures on it, and the weight and space required to carry it—and that means we simply end up with more things that we need to describe. Without ways to manage this process, we can lose the advantages that digital media provides. Traditional library science principles of description were designed to deal with traditional documents—as Svenonius describes them, works that could be associated with “a discrete physical object.” However, digital documents break this frame. Although they are physical in that they take up space in the form of bits and bytes, they can be “unstable, dynamic, and without identifiable boundaries” (Svenonius, 2000, p. 13). The problems fall mostly into three major categories. We’ve already discussed the semantic gap, which occurs when a computer process encodes or describes information in a way quite different from the way humans would describe the same information. That’s a significant issue with multimedia. (Have you ever tried to open a music file in a text editor? The resulting string of apparent nonsense would make anyone seeking to describe what that file was “about” give up.) Another challenge is the sensory gap: the difference between an object and the way a computer can sense, interpret, and describe it. And the massive amount of media that can be produced now by professionals and amateurs alike leads to a proliferation problem, where there’s just so much to describe and often no simple or consistent way to do it quickly.

Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

3.3.2. History of Describing Multimedia Instances The problems of describing multimedia instances aren’t all new. Museum curators have been grappling with them since they first started to collect, store, and describe artifacts hundreds of years ago. Many artifacts may represent the same work (think about shards of pottery that may once have been part of the same vase). The materials and forms don’t convey semantics on their own: Without additional research and description, we don’t know anything about that vase; it doesn’t come with any sort of title page or tag that identifies it with a 9th-century Mayan settlement. And since museums can acquire large batches of artifacts all at once, they have their own version of the proliferation problem. One approach to these problems was first codified by the German art historian Erwin Panofsky. In his Studies in Iconology (1939, p. 5-9), Panofsky declared that there are three primary levels of understanding necessary for properly describing an artistic work or museum artifact: • Primary subject matter: At this level, we describe the most basic elements of a work. The painting The Last Supper, for example, might be described as “13 men having dinner.” • Secondary subject matter or identification: Here, we introduce a level of basic cultural understanding into a description. Someone familiar with a common interpretation of the Bible, for example, could now see The Last Supper as representing Jesus surrounded by his disciples. • Intrinsic meaning or interpretation: At this level, context and deeper understanding come into play—including what the creator of the description knows about the situation in which the work was created. Why, for example, did this particular artist create this particular depiction of The Last Supper in this way? Panofsky posited that the role of the professional art historian was key at this level, because they were the ones with the education and background necessary to draw meaning from a work. In other words, Panofsky saw the need for many different types of descriptors—including physical, cultural, and contextual—to work together when making a full description of an artifact. Panofsky’s principles have persisted. But these days, there are yet more types of media needing to be described, as well as more computational processes that can help us build the descriptions. We’ll look at some examples in the following section.

3.3.3. Modern Methods of Describing Multimedia Instances The way we describe non-text instances depends greatly on the individual properties of those specific instances, as part of the challenge is that each type of multimedia or multidimensional instance can have wildly different properties. In this section, I’ll describe some of the approaches people have taken to describe different types of multimedia or multidimensional instances. Music Some parts of describing a song aren’t that different from describing text: You might want to pull out the name of the singer and/or the songwriter, the length of the song, or the name of the

Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

album on which it appears. But what if you wanted to describe the actual content of the song? You could write out the lyrics, but describing the music itself requires a different approach. A DJ, for example, might care greatly about the beats per minute in each song. If you’re making a playlist for a road trip, you might be seeking songs that you’d describe as “good for driving” — though you’d have to figure out what “good for driving” means first, which is a highly subjective description. And if you’re looking for recommendations for new bands, you might want to know how to find music that’s somehow like music you already know you love. Several people and companies working in multimedia have explored different processes for how songs get described. On the heavily technological side, software applications such as Shazam and Midomi can create an “audio fingerprint” from a snippet of music. Audio fingerprinting renders a digital description of a piece of music, which a computer can then interpret and compare to other digital descriptions in a library. On the other hand, the online radio service Pandora uses music experts, not computers, to do much the same thing. The company employs an army of coders, including trained musicologists, who listen to individual pieces of music and determine which words from Pandora’s highly controlled vocabulary for musical description apply to a given song. The result is Pandora’s “Music Genome,” an algorithm that ultimately recommends songs for its users by stripping down the songs they say they like to their component parts and suggesting, for example, more songs with “driving bass” or “jangly guitars” (Walker, 2009). Images Much as with music, certain properties of images are easy to describe — their size, whether in inches or in kilobytes, or the date they were taken. But again, describing the content of an image in a way that people and/or computers can understand is a trickier proposition. Google’s Image Labeler game1 is one effort under way to help describe the content of a photo. Operating under the assumption that different people use different words to describe the same thing—while also banking on the idea that people will eventually “match” or settle on a mutually acceptable description—the game randomly pairs people together to suggest labels or tags for an image. Typically, the obvious choices are removed from contention, so a photo of a bird against a blue sky might already strike “bird” and “sky” from the set of acceptable words, leaving users to suggest words such as “flying” and “cloudless.” Players earn points for working their way through the game, but the real benefit is in providing data to Google that helps determine what photos to return when someone uses the company’s image search engine. The Library of Congress has a similar project under way on the photo-sharing site Flickr, which allows registered Flickr members to add tags and descriptions to the images from the LOC’s archive, posted on Flickr. Other services attempt to automate the creation of a description for an image. Much the same way music files can have an “audio fingerprint,” images can have an “image signature” or be measured on their “visual similarity” to one another. The company Idée has devoted itself to developing a variety of image search algorithms, which use image signatures and measures of visual similarity to return photos similar to those a user asks to see based on “hundreds of image attributes such as colour, shape, texture, luminosity, complexity, objects, and regions.”2 The                                                          1 http://images.google.com/imagelabeler/  2 http://labs.ideeinc.com/faq.html#visual 

Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

company’s Multicolr search, for example, returns a set of stock photos with similar color combinations to the ones selected dynamically by the user. Video Video is yet another area where work to create descriptions to make search more effective is ongoing. Identifying the content of a video currently takes a significant amount of human intervention, though it’s possible that image signature-matching algorithms will take over in the future. One organization that sees a future in assembling better descriptions of video content is the United States’ National Football League (NFL), whose vast library of clips can not only be used to gather plays for highlight reels and specials but can also be monetized by pointing out when key advertisers’ products appear on film. Currently, labeling the video requires a person to watch the scenes and tag elements of each frame, but once those tags have been created and sequenced along with the video, they can be more easily searched in computerized, automated ways (Buhrmester, 2007). YouTube is using a similar process to introduce targeted ads into its service. Knowing when two characters on a teen soap opera are discussing their hair, for example, could help the service know when to insert a small advertisement for shampoo. “Smart things” and sensor networks As we discussed in section 2.4, sensors are now making all sorts of objects “smarter,” capable of reporting their status, their location, or other important descriptive data. Many applications of “smart” sensors are still in their infancy, and that makes them interesting to study for how descriptions can be created automatically and processed (automatically, by people, or both) down the line. Some sensors create relatively simple descriptions: A pass that registers a car’s location at a toll booth, for example, doesn’t need to do much besides communicate that the particular sensor has arrived at some location. In essence, it acts as an identifier. But the information created or tracked by sensors can be more complex, as well. Some sensors can calculate location using GPS coordinates and satellite tracking, while others can take readings of temperature, pollution, or other measures. These readings can be taken on their own or combined into richer and more detailed descriptions of some thing or event. The tradeoffs in creating these descriptions likely sound familiar by now: More descriptors can create a fuller and more accurate picture of the world, but they require more processing power to collect the necessary information and render it into a meaningful form. One interesting service that uses sensors to create descriptions of location is the NextBus transportation tracking service, which aims to tell transit riders exactly when vehicles will be arriving at particular stops. NextBus uses sensors to track the GPS coordinates of buses and trains, then compares that to route and information from transportation providers and estimates the time it will take for a vehicle to arrive at some selected location. To offer information that will be useful to riders, NextBus must figure out how to describe the location of a vehicle and the distance between that location and some intended target, as well as creating descriptions of transit routes (name, number, and/or direction of travel) and particular stops along the route. In some areas, NextBus incorporates multiple descriptors for a given stop by allowing users to search by route, by location, or by an ID number assigned to the stop.

Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

3.4. Who Chooses Descriptions? One of the greatest challenges of description is coming up with the right words. If you’re describing a thing for your own purposes—say, you’re labeling photos on your own private Flickr account, where you are the primary retrieval target—you’re free to choose whatever description makes sense to you. But as soon as other people come into the picture, the rules and goals of the description change. Instead of just trying to describe information for your own purposes, now you have to take the needs of others into account. As we discussed in the chapter two section on naming, one step in deciding how to name something can be the development of a controlled vocabulary — a set of words, definitions, and rules of use that cover all the possible terms for naming something in a given domain. The principles of controlled vocabularies can also apply to descriptions in a broader sense, and in this section, we’ll discuss some of the ways those vocabularies get developed. Of course, plenty of description happens outside the boundaries of a controlled vocabulary — for better or worse. We’ll examine that here too.

3.4.1. Describing is Challenging and Contentious Like naming, describing an instance can be challenging and contentious. Our cultural and individual experiences often leave us with strong opinions on what description is the “right” one for a given situation, as do our political leanings, business goals, and even our moral beliefs. If I’m an automobile manufacturer, for example, I may want to describe my sport utility vehicles as “light trucks” so I can be held to less-strict government emissions standards. If I’m an environmental policy lobbyist, however, I would want even more vehicles to be described as cars so they would all have to meet the strictest standards.

3.4.2. People Choose Descriptions All day, every day, we decide how to describe things. We figure out how to communicate that our office is in “the first building on the left,” that our photo is a picture of “three flowers in the back yard,” that the dinner we chose at the restaurant was “a little too spicy.” If we’re authors, we decide how to describe our own works by giving them titles and subtitles, topic sentences and abstracts. If we’re business owners, we might decide what our company is about by giving it a name that reflects the qualities we want to convey to customers and potential business partners. When we’re describing information for our own purposes, our own words are often all we need. If we’re cataloging our own files, or organizing our own pictures, or deciding how to file our recipes that only we will ever cook, we can choose whatever descriptions make the most sense to us. If we want to give our pictures specific titles such as “July by the Golden Gate Bridge,” that’s just fine. If we want to pick something less obvious like “San Francisco Picture No. 400”, that’s fine, too. Problems can arise, though, when those descriptions that we chose need to make sense to other people, or when the descriptions of others need to make sense to us. As we discussed in chapter two, people inherently use a huge variety of words to describe the same things and concepts, and only a relative few of those words are used with any frequency. As we mentioned

Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

in Chapter 2, George Furnas and his colleagues termed this issue the vocabulary problem after seeing the number of different terms people applied to the same set of documents or actions. One solution to the vocabulary problem is to create and impose a controlled vocabulary, but that’s something we rarely do on our own. Creating a vocabulary involves many decisions about the intended audience, the scope, the scale, and the granularity of the descriptions it will include. Making those decisions on an individual basis can be difficult. Some suggest that, in our increasingly digital world, a more realistic solution is to allow everyone to assign descriptors — or “tags,” in web parlance — that make sense to them, without any intervention to create a central controlled vocabulary. As David Weinberger, a key proponent of this point of view, writes, “We can confront the miscellaneous directly in all its unfulfilled glory. We can do it ourselves and, more significantly, we can do it together, figuring out the arrangements that make sense for us now and the new arrangements that make sense a minute later” (2007, p. 22-23).

3.4.3. Information Professionals Choose Descriptions On the other end of the spectrum, we have the more rigorous descriptions chosen by librarians and other information professionals. Rather than abide by our messy and inconsistent individual descriptions, professionals set standards for what makes an effective description of an instance and then work to make sure all descriptions meet those standards. The specific standards can vary. Within library science, several different descriptive schemes — Library of Congress, Dublin Core, etc. — exist to set rules for what makes a good description. But just as some believe that allowing the entire world to create descriptions could lead to chaos, there are some who say that the more standardized, rigorous, professional way of choosing descriptions has not done enough to keep up with the changing times and the quantity and type of information we now want to catalog. As Denton writes, “There are classification schemes and subject heading systems. There are thick manuals on how to catalog serials. There are thick manuals on library software that implements the standards. There are thick manuals on everything. Special fields like medical or legal librarianship have their own rules. . . . It’s hard to keep up with all that, but the real problem is that all those rules can’t keep up with what is happening around us” (Denton, 2007, p. 50).

3.4.4. Businesses Choose Descriptions Businesses choose descriptions for several different reasons. They may want to describe particular products or services in a way that will resonate with potential customers. And they also may want to describe their own business processes in a way that will allow them to mesh more seamlessly with collaborators. In early 2010, the financial services company Vanguard set out to create a description of the concept of financial security. Perhaps unsurprisingly, the word they chose was “vanguarding.” According to an advertising manager who worked on the campaign, the financial services market is full of companies “claiming this territory of ‘We care about you,’ ‘Trust in us’” (Elliott, 2010, para. 24). By choosing to describe their services as “vanguarding” — and define a meaning for that word in the process — the company sought to become memorable for something different.

Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

On the flip side, Sony has come under fire in recent years for not embedding relevant or even decipherable descriptions within the names of its products. As Wired magazine lamented, the old-school Sony Walkman suggested some concept of what the device could do merely with a descriptive name. But the company has gotten away from that: “In the market for a 3-D TV? Would you prefer an XBR-60LX900 or an XBR-46LX900?” (Greenfeld, 2010, para. 18). The Wired writer contrasts this not only with Sony’s own product history but also with the Kindle, Flip, and Wii, “simpler, cheaper gizmos” with easier-to-understand names. Choosing descriptions for a product can sometimes veer close to branding. Think about how Apple computers have been marketed—as cool, sleek, revolutionary, well-designed. Now think about the marketing for PCs. Key ideas are inherent in each of those descriptions, whether or not the words are ever directly associated with a particular product. Other stories are less dramatic but no less important. Choosing descriptions as a single business can be a lot like choosing descriptions as a single person: as long as you choose descriptions that work for you and your internal documents and information processes, all is well. But as soon as you need to expand those descriptions to other businesses — to exchange information, for example — then the descriptions take on a different level of importance. See chapter 10 for much more on this.

3.4.5. Governments and Interest Groups Choose Descriptions Governments and governmental bodies can have a vested interest in how various products, services, and concepts get described. Their attitudes toward descriptions can vary depending on legal or safety regulations, political goals, or other potential ramifications of choosing one description over another. In one case, a wing of the U.S. government sought out cultural descriptors to help pilots navigate their flight paths and landings. Until the ‘70s, the Federal Aviation Administration used meaningless combinations of letters for the five-letter codes But starting in 1976, the “fixes” started to be named after things pilots might know about certain cities: MICKI, MINEE and GOOFY are the names for navigation points near Orlando, Florida, where Walt Disney World is located, while BRBON marks a spot in Kentucky—the home of Bourbon County—and BOSOX tells pilots they’re getting close to Red Sox country (McCartney, 2006). In another case, though, a different governmental agency made a point of removing familiar touchstones from its own description process. The Food and Drug Administration, which must approve the names for all new medicines that hit the U.S. market, has rejected names that are “semantically suggestive” — in other words, those that might give the appearance of a drug promising more than what it can actually do. The osteoporosis drug that came to the FDA with the name “Bonviva” — presumably chosen by the manufacturers to reap the benefits of “bon” = good and “viva” = life — ended up with the slightly nonsensical and less evocative name “Boniva” (Matthews, 2006).

3.4.6. Computers and Other Automated Processes Choose Descriptions Computers frequently assign default descriptions to information instances. Digital cameras contain images tracked as DSC_024 or IMG000045. When I start up Microsoft Word and open a blank document, it’s automatically assigned the name Document 1. And if I import a music file

Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

into iTunes without giving it a name, it gets a default, typically something like Track 002. Sometimes these computer descriptions are incredibly difficult for people to parse. IMG000045, for example, might mean very little to you offhand; you might know it’s a picture, but you might not have any idea where or when you took it — or what it was a picture of — without looking at the photo itself. And if I open a bunch of new document windows in Word, I very quickly have a hard time remembering what I was typing in Document 3 and what was happening over in Document 7. In situations like this, the descriptions assigned by computers start to drift toward the semantic gap, the divide between the names people assign to things and the ones processes assign. Sometimes, though, computer descriptions can actually be more understandable, cutting through all the fluff and imprecision of human language. Take, for example, the basic idea of color. What I call “blue” might be your “aqua” or “teal.” But in HTML, each color gets assigned a specific code, based on the number of red, blue, and yellow in it. So, “true blue” in HTML parlance is #0000FF — and one person’s #0000FF will forever be the same as another’s. (Granted, deciding what #0000FF would equate to probably did involve human judgment at some point — but at least it set a benchmark for what we mean when we want to see blue on the internet.) Computers and computer processes also help to decide which words in a description are useful and which aren’t. “Stopwords,” for example, are the words (typically prepositions, helper verbs, and the like) that get filtered out of a search query because they’re too insignificant or occur too frequently to be useful for distinguishing one document from another. (Of course, any description “chosen” by a computer program is really chosen by a person—the programmer responsible for that code—but for the purposes of this discussion, let’s leave that aside.)

3.5. What Makes a Good Description of an Instance? With all of the challenges of description, how do we know that the particular description we’ve chosen for an instance is even a good one? This section will explore the way different disciplines have approached the answer to this question over the years.

3.5.1. No Descriptive Language is Perfect for Every Situation As we’ve been discussing throughout this chapter, description requires thinking through a number of tradeoffs. How much time do you want to spend organizing your information — and how much time are you willing to spend trying to retrieve it later? Will a simple description suit your needs, or do you need something quite complex and granular to distinguish one item from another? Do you need the semantic meaning that a human description can provide, or is the mechanical code applied by a computer enough? No descriptive language is perfect for every situation. You wouldn’t want to go through the trouble of creating a controlled vocabulary if something less rigorous would suffice; on the flip side, you also wouldn’t want to use imprecise natural language descriptions in a situation where specifics really mattered.

Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

The lesson is to always fully think through the situation. Who’s the audience? What’s their goal? What’s yours?

3.5.2. Traditional Principles In The Intellectual Foundations of Information Organization, Svenonius (p. 68) suggests a number of principles that determine whether a description is sufficient for bibliographic purposes. These principles shape the design of the language used for organizing information, and adherence to them determines the success of a description. • • • • • • • •

User convenience: choose descriptions with the user in mind Common usage: choose words that the majority of users will understand Representation: base descriptions on the way an entity describes itself Accuracy: faithfully portray whatever it is that’s being described Sufficiency and necessity: descriptions should have enough information to achieve their objectives and exclude information that’s not necessary for meeting these objectives Significance: include only those elements that are bibliographically significant Standardization: standardize descriptions to the extent possible Integration: base descriptions for any type of material on a common set of rules

These principles introduce plenty of questions of their own, questions that it’s essential for an information professional to be able to answer when generating a description. Who is the user? What is the user’s language, and how will the user understand the words chosen for a description? What information is absolutely necessary for a description to be understood? Svenonius points out that many of these principles are vague and open to interpretation. But they serve as a useful starting point for asking whether a description will work for a given situation.

3.5.3. Business and Standards-Related Principles Businesses – or, really, any two entities that have to share information with each other – often have specific requirements for what makes good descriptions to foster interoperability and ease information exchange. As we mentioned in chapter 2, and will discuss in greater depth later in this book, ensuring that various parties can understand each other’s information is essential to commerce, government, and the functioning of many other complex systems. In Document Engineering, Glushko and McGrath propose a set of generic requirements for documents, which can be applied in some way to nearly every situation where a document or information component is being described (p. 252-253). Many of these generic requirements can be used as tests for evaluating description in the business world. Some echo Svenonius’s principles, while others add new layers to the evaluation of a description. The list includes: • Completeness: Ensure that a document contains all the information it should or that its recipient (person or application) expects. • Accuracy: Ensure that every piece of information in a document is correct. • Usability: Present information in a format or medium that is easy to use and understand by its intended users.

Chapter 3: Describing Instances  •

 

Last updated: September 9, 2010 

Identifiability: Ensure that the design or appearance of a document signals that it comes from our organization or company; also called branding of the information.

Glushko and McGrath note that these requirements – among others they mention as fundamental – may not all apply to every situation, nor are they necessarily compatible with each other. But again, the most important thing for an information professional is to keep the requirements in mind. Talking to stakeholders to find out what they need to know can help determine what would make a description complete, accurate, usable, and identifiable. Document Engineering also mentions standards compliance as a requirement. We’ll discuss standards more in the coming chapters, but for now, suffice to say that different organizations set different standards for how information is to be described. The key thing to know is that descriptions may be required to follow a particular format or conform to a particular standard, based on the rules of a given organization and the standards to which it subscribes.

3.5.4. Computational Principles As we discussed earlier in this chapter, people—both expert and amateur—create many of the descriptions we use. But mathematical principles can also help to determine the best descriptions for certain situations by analyzing which properties are the most unique and essential to describe, which words will be the most precise, and which things are most like and unlike each other. Later in this book, we’ll discuss some of these methods—including dimensionality reduction and latent semantic indexing—in greater depth. For now, though, consider that computational methods can help us determine: • How many descriptors we need. Computers can quickly look over a large set of data and determine which properties are common to all members of a given set. Let’s say every time something has been described as “red,” it’s also been described as “big.” In that case, we don’t need to describing something as both “big” and “red.” The two terms just go together. • Which descriptors are the most important. Given a lot of time, people might be able to make the connection in the previous example by noticing that all things that are either “big” or “red” are, in fact, both big and red. But if we’re only going to use one of those words in the description, should it be “big”? Or is the better descriptor “red”? Computational techniques can be used to analyze large amounts of information and determine which individual descriptors are the most important to attach to some instance in a particular collection. For example, algorithms can be written to determine which words are most likely to appear together in certain kinds of documents. • Which things are most like or unlike each other. Clustering algorithms can help to determine when instances share key descriptive properties. If you’ve ever taken a MyersBriggs personality test, you’ve experienced this in action: The test aims to take a range of human behaviors and group them together into like categories (introvert or extrovert, sensing or intuition, etc.), reducing all people to one of 16 descriptions. Your MyersBriggs type may not describe you perfectly—but the goal is that every INFJ will be more like every other INFJ than like an ESTP. This type of evaluation will come in to play in chapters 5 and 6, when we’ll discuss classification and categorization.

Chapter 3: Describing Instances 

 

Last updated: September 9, 2010 

Sources Buhrmester, J. (2007). NFL Films' Exhaustive Archive is Rushing Into the Digital Age. Wired. Retrieved from http://www.wired.com/culture/lifestyle/magazine/15-10/ps_nfl. Denton, W. (2007). FRBR and the History of Cataloging. In A. Taylor, Understanding FRBR: What It Is and How It Will Affect Our Retrieval (pp. 35-57). Westport, Connecticut: Libraries Unlimited. Foust, J. (2006). Demote Pluto, or demote “planet”? The Space Review, August 28, 2006. Retrieved from http://www.thespacereview.com/article/692/1. Glushko, R. J., & McGrath, T. (2005). Document engineering: analyzing and designing documents for business informatics and Web services. MIT Press. Greenfeld, K. T. (2010). Saving Sony: CEO Howard Stringer Plans to Focus on 3-D TV. Wired. Retrieved from http://www.wired.com/magazine/2010/03/ff_sony_howard_stringer/. Martin, A., & Krauss, C. (2009). Pork Industry Fights Concerns Over Swine Flu. The New York Times. Retrieved from http://www.nytimes.com/2009/04/29/business/economy/29trade.html?_r=1. Panofsky, E. (1939). Studies in iconology : humanistic themes in the art of the Renaissance. New York, NY: Oxford University Press. SarcMark. Retrieved from http://02d9656.netsoljsp.com/SarcMark/modules/user/commonfiles/loadhome.do. Svenonius, E. (2000). The Intellectual Foundation of Information Organization. The MIT Press. Retrieved from http://www.amazon.com/dp/0262512610.   Tillett, B. (2003). What is FRBR? A Conceptual Model for the Bibliographic Universe. Turnbull, G. (2009). A Common Nomenclature for Lego Families . The Morning News. Retrieved from http://www.themorningnews.org/archives/opinions/a_common_nomenclature_for_lego_families. php. Walker, R. (2009). The Song Decoders at Pandora. The New York Times. Retrieved from http://www.nytimes.com/2009/10/18/magazine/18Pandora-t.html. Weinberger, D. (2007). Everything Is Miscellaneous: The Power of the New Digital Disorder (p. 288). Times Books. Retrieved from http://www.amazon.com/Everything-Miscellaneous-PowerDigital-Disorder/dp/0805080430.