Oranges and peaches: understanding communication accidents in the reference interview

Patricia Dewdney and Gillian Michell

Introduction by Gail Schlachter, president of Reference Service Press

The Reference Service Press Award, $1,000 and a plaque donated by Reference Services Press Inc., is given annually to the author(s) of the most outstanding article published in RQ, the official journal of the Reference and User Services Association, during the preceding two volume years. (RQ is now called Reference and User Services Quarterly.) The 1998 award was given to Patricia Dewdney and Gillian Michell for their article “Oranges and Peaches: Understanding Communication Accidents in the Reference Interview”(vol. 35, no. 4, Summer, 1996, pages 520-536). 

Dewdney and Michell “take these ‘communication accidents,’ which reference librarians often talk about in a purely anecdotal way, and apply an academic, linguistic analysis to them,” said David Null, chair of the Reference Service Press Award selection committee. “The authors classify these ‘illformed queries’ into four main categories and offer ways that librarians can ‘avert or repair such accidents’.”

The reason their article was chosen for the award was because it is one that all types of reference librarians, or indeed any person who deals with the public, can relate to and learn from.

The following story recently appeared on a newsgroup for reference librarians:

The young undergraduate was standing in line at the Graduate Collections desk. When he reached the desk everyone within earshot found out why. “I just can’t find this book,” he blurted out. “I have to read it. I tired to find it on my own and I can’t!!!! Those idiots over at the undergraduate desk were no help. Can you help me?” 

The librarian replied in a soothing voice, “Sure! Can you give me a title?” 

“It’s called Oranges and Peaches!” the student replied. “I’m at my wits’ end!”The librarian calmly typed “Oranges and Peaches” into the computer catalog, then frowned. The library had no book with that title. The librarian asked gently, “Do you have an author for me?” 

The question seemed to trigger more anxiety and frustration in the student. “Can’t you find it? How could you not have this book?” he asked, his voice rising. “T.A. and the professor said you’d have hundreds of copies of this book!!!!” 

The reference librarian persisted.”Perhaps you could tell me the author?” The young man dived into his backpack, searching through papers for a seemingly interminable time, to the growing annoyance of people waiting behind him. Finally, he found a scrap of paper. “Charles somebody,” he said triumphantly. 

The young man’s face fell as the librarian apologetically that he couldn’t look up a book under a first name. The librarian added, “I’m sorry, but you’ll have to go back to your instructor and get a last name.” 

The student’s despair turned to rage. “I cannot believe this,” he exploded. “What kind of library is this? T.A. swore you’d have hundreds of copies. This book is meant to be legendary! Professor said it’s the ‘Bible of evolution’!” 

Comprehension dawned. “Could it be On the Origin of Species by Charles Darwin?” the librarian asked. “We do have lots of copies of that.” 

“Yeah, that’s it!” said the relieved undergraduate.

This story, taken from LIBREF-L, has been edited for brevity. Variants have shown up on other newsgroups. Considering the thousands of brief encounters that reference librarians have with library users, it is not surprising that periodically they have communication accidents like the “oranges and peaches” story. Every public librarian can readily recall a similar story: the man who insisted on seeing the book Down on Broad Street (which turned out to be Dun and Bradstreet), or the woman who wanted books on “gynaecology” for tracing her family tree.

Communication accidents in ordinary conversation have been extensively studied by researches from various disciplines including linguistics, education, cognitive science, and artificial intelligence. (For an overview of this work, see Ronan G. Riley, ed. Communication Failure in Dialogue and Discourse: Detection and Repair Processes (Amsterdam: Elsevier Science, 1987), 99-120.) One reason for the recent focus on conversation failures is the amazing ability of human beings to repair their communication accidents, and the implications this ability may have for automated question-answering systems. (Ibid., v.) As Ringle and Bruce state,

[human] conversation failure, in fact, appears to be the rule rather than the exception. The reason that dialogue is such an effective means of communication is not because the thoughts of the participants are in perfect harmony, but rather because the lack of harmony can be discovered and addressed when it is necessary. The nature of this cooperation is what we would like to understand. 

H. Ringle and Bertram C. Bruce, “Conversation Failure,” in Strategies for Natural Language Processing, ed. Wendy G. Lehnert and H. Ringle (Hillsdale, NJ: Lawrence Erlbaum, 1982), 203-221, 224.

In the reference transition, the librarian must have a clear, complete understanding of the user’s information need before a satisfactory answer can be given. Often the question must be negotiated through a reference interview, where the librarian will attempt to clarify, expand, and perhaps repair the query as it is initially presented by the user. (Previous studies have suggested that about 50 percent of the initial queries presented by users require a reference interview. See, for example, Jo Lynch, ”Reference Interviews in Public Libraries,” Library Quarterly 48 (April 1987): 119-142, and Patricia Dewdney, “The Effects of Training Reference Librarians in Interview Skills: A Field Experiment” (Ph.D. diss., Univ. of Western Ontario, 1986) Some initial reference questions may be downright puzzling, as in the “oranges and peaches” story, and yet librarians seem to be able to resolve them. Such questions can be included in what some researchers have termed “ill-formed queries,” or natural-language questions presented in a form that, for various reasons, cannot be presented to the system until the information need has been sorted out and the question repaired. (For example, see Ralph M. Weischedel and Norman K. Sondheimer, Meta-rules as a Basis for Processing Ill-formed Input,” in Communication Failure in Dialogue and Discourse: Detection and Repair Processes, ed. Ronan G. Reilly (Amsterdam: Elsevier Science, 1987), 99-120.) 

This article concentrates on the occurrence in the reference process of one type of ill-formed query, relating it to what Ringle and Bruce have called input failures and model failures. An input failure occurs when the listener is unable to obtain a complete, or at least coherent, interpretation for an utterance; a model failure occurs when the listener is unable to assimilate inputs into a coherent belief model as intended by the speaker. (Ringle and Bruce, Strategies for Natural Language Processing, 208)

In this article, we present a taxonomy of input failures and some simple model failures with an explanation of the linguistic reasons for these failures, and a commentary on the social and environmental context in which these communication accidents occur. 8 We also present some strategies and interview techniques that reference librarians can use to prevent or repair many of these failures. 

Sources of the data
The examples given in this paper are all real. They have been taken from a database of several hundred records of “reference interviews gone wrong” collected from academic, public, and special libraries in six ways: reports from librarians attending workshops in communication skills; in-depth interviews with experienced librarians; transcripts of actual reference interviews; postings on LIBREF-L, and Internet newsgroup for reference librarians; and anecdotes reported to the researchers. 9 Examples were taken from the first two sources, since these data were collected specifically for the purpose of documenting communication accidents. These involved (1) reports from practicing librarians who participated in workshops on communication skills, and were asked to recall occasions when the initial question presented by the user turned out to be something quite different; and (2) transcripts of twenty-six in-depth interviews with experienced public library reference staff who were asked to recall a recent transaction in which a misunderstanding developed from the way in which the user initially presented the query, and were then asked to describe each cognitive event through which the librarian came to an accurate understanding of what the user really wanted. 10 

It is not our purpose in this article to investigate either the frequency of communication accidents or the extent to which our findings our typical. We are interested simply in identifying and analyzing as many different types of communication accidents as possible. Therefore, instead of counting all examples of a given type, we worked with the data until we had reached a saturation point – i.e., where no new categories could be identified – and then developed a taxonomy. Because the data were collected systematically over a number of years from practicing librarians, we are confident of the authenticity of the examples. However, since most examples came from librarians’ self- reports rather than from actual recordings of the conversation between librarians and users, these accounts are subject to the usual limitations of human memory and perceptual biases. The validity of the examples and their meaning can only tested only through reactions of informed readers to this paper. 

Characteristics of words
Communication accidents arise from the misinterpretation of one or more key words. In considering the kinds of communication accidents that happen at the reference desk, it is helpful to understand first how linguistics have analyzed the makeup of words. As speakers of a language, when we know a word and how to use it, we know four kinds of information about it, information that is technically called phonological, syntactic, semantic, and pragmatic. 

Phonological information tells us how to pronounce the word and its different forms. It tells us that the second e in telegraph has a different sound from the second e in telegraphy – the first more or less like “uh” and the second like the short “eh” in met. We can often recognize words that are pronounced in a systematically different way by speakers of other dialects of English and by non-native speakers of English. For instance, we know that Canadians pronounce the sound written ou in words like shout in a way that distinguishes Canadians from Americans, and we recognize that Francophones often pronounce the English “th” sound as “t” or “d”. Even our own pronunciation can vary from one occasion to another in minor ways that normally don’t affect comprehension. Further, if we are literate, we know the written form of the word and often get clues to identifying it from the spelling that we might miss if we just heard it: a word pronounced “fie-la” is suddenly recognizable as a biological term when we see it spelled phyla. 

Syntactic information that we know about a word includes the grammatical category to which it belongs – what we used to call parts of speech. In addition, we know syntactic facts that tell us how a word can be used grammatically in sentences. Is it a noun? Can it be pluralized? Is it a proper noun? Is it a verb? Does it take a direct object? If not, “smiled Susan” will be recognized as an ungrammatical use of smile, and we would correct it to “smiled at Susan”. We also know semantic information – information about meaning – when we know a word. In addition to knowing the concept or thing in the world to which the word refers, we know what other categories of meaning apply to it. Is the referent concrete or abstract? Animate or inanimate? Human? Without this knowledge we would produce anomalous sentences, such as “The rock whom I sat on…”
Finally we know pragmatic information about a word; that is, aspects of meaning that have to do with the context or real-world situation in which it is used. Knowing whether a word is slang or offensive or obsolete is part of pragmatic knowledge. The word frog can refer to an amphibian or to a French-speaking person, but the latter meaning carries social meaning as well, since it is used disparagingly. Distinguishing such a word from others that sound like it or almost like it often depends on hearing the word in context. 

How do these characteristics of words relate to communication accidents at the reference desk? An example will illustrate how these distinctions allow us to describe what is going on. A librarian had spent the day weeding the sports section of the reference collection. He was approached by a user who asked for books on “soccer tees”. Unfamiliar with the concept, he was prepared to provide the user with books on soccer when the user elaborated on his original request, adding “the Greek philosopher.” In this example, the librarian’s previous focus on a particular subject area predisposed him to interpret the minor sound difference between “soccer” and “socra-“as the former. That is, the pragmatics of his situation led him to identify the phonology of the user’s word in a particular way; in this instance, it was the wrong way. 

The examples examined in this paper can be categorized as failures that occur for phonological, syntactic, semantic, or pragmatic reasons, but since many of our examples involve more than one of these, we have chosen instead four broad categories that illustrate increasing levels of difficulty for the librarian, ranging from those that are very simple, easily recognized and corrected, to those that are linguistically more complex, often involving pragmatic or model failures, and this more difficult to recognize, explain, and correct. Briefly, the categories are as follows: 

(I) No harm-done accidents, or recognized errors;
(II) Unrecognized librarian-originated accidents, where the librarian does not interpret the initial query in the way that the user intended;
(III) Secondhand communication accidents, which are based in previous errors of interpretation on the part of the user and repeated to the librarian; and
(IV) Creative reconstruction, where the user “reconstructs” the meaning of a forgotten word or phrase but the result does not match the original words. 

Analysis of four types of communication accidents
Class I: No-harm done accidents
Class I examples are straightforward errors of hearing or understanding that are recognized as such by the librarian and corrected immediately. For example, an input failure sometimes occurs because of environmental noise (such as a lawnmower operating outside the library window) or because the user speaks too softly, as in the following transcripts of audio-taped reference interviews.(The method used for this study was the time-line interview, adapted from a technique developed by Brenda Dervin and described in Brenda Dervin, “From the Eye of the User: the Sense-making Qualitative-quantitative,” in Qualitative Research in Information, ed. Jack D. Glazier and Ronald R. Powell (Englewood, CO: Libraries Unlimited, 1992), 61-84.)

User: ((Where do you keep the computer books?))
Librarian: Sorry?
User: Computer (( )).
Librarian: Computer books? Anything on computers?
User: No, graphics.

In such cases, the librarian knows she hasn’t heard the user’s words clearly, so she immediately takes steps to repair the accident, usually by asking the user to restate the question through a convention such as “Sorry?” The next example is a variation on acoustic failure, since the librarian seems to cope well with most of what the user says until he mentions an unfamiliar name.

User: um- (( )) comic book character from (( )) British comics just after the war called Billy Bunter.
Librarian: Billy --?
User: Bunter. Billy Bunter.

The first example corresponds to Ringle and Bruce’s category of acoustic or perceptual failure, whereas the second is a type of lexical failure, where the word may be hear but fails to produce any kind of interpretation. (Ringle and Bruce, Strategies for Natural Language Processing, 208) Both are input failures.

Class II. Unrecognized librarian-oriented errors
Class II examples are based on (initially) unrecognized errors of interpretation on the part of the librarian. These kinds of errors vary in their origin and nature, and fall into three subcategories.

Subcategory IIa. Pronunciation variants.
Here the source of the error is phonological: something about the pronunciation of the word is sufficiently off the norm that the librarian interprets the word to be something different from that intended. This can arise from language or dialect differences. In some of these cases, the librarian may be cued to the potential for a misunderstanding through nonverbal signals. For example, she may infer from the user’s appearance that he may not speak English fluently.

A classic example was reported by a university librarian who described the user as “Asian”. To the librarian, it sounded as if the user was asking for “Volume Nine,” which he repeated after she asked several times, “Volume 9 of what?” What he wanted was the business reference, Valueline. Naturally, the exchange was frustrating and embarrassing for both. In a similar example, the librarian thought the user (who had a French accent) asked for information on “bird control,” but what he really wanted was information on birth control.

A problem arising from dialect is described by a librarian in St. John’s, Newfoundland, where there are many different dialects of English. The user’s request for information on “gaffix” made no sense to the librarian. The subject was eventually discovered to be Guy Fawkes. In another Canadian public library, the librarian thought the user said, “Do you have any books on how to ride a whale?” which turned out to be “how to write a will.” This misunderstanding also arises from differences in dialect. Canadians, and some other dialect speakers, pronounce the vowel in “ride” and “write” differently, whereas in other dialects they are the same. A Canadian hearing someone with the other dialect say “write” would be inclined to interpret the word as “ride,” and then for pragmatic reasons would be inclined to hear “whale” rather than “will” (since you can’t ride a will). The concept of riding a whale, while not an impossibility, is unusual enough that the librarian would want to ask further questions. Similarly, a question that seemed initially to be about hairdressers in Saudi Arabia turned out to be “addresses in Saudi Arabia,” perhaps because the user spoke a dialect of English in which the post-vocalic r is not pronounced.

In a more unusual example collected in a California library, the librarian thought the user said, “Do you have something on your chest?” Although this is an utterance that is syntactically and semantically correct, the librarian suspected communication failure because this is not the sort of thing users generally say to librarians; that is, the utterance fails on pragmatic or contextual grounds. The librarian reported the user was Hispanic, and when she asked him to repeat his question, she realized it was “What do you have on how to play chess?” Another example of a communication accident involving Spanish occurs in a request for information on “laws,” which was heard by the librarian as “lace”. This can be explained if one knows that (a) in Chicano speech, Spanish and English words are often mixed, (b) the Spanish word for laws is pronounced as “lay-es”, and finally, (c) the ending es is often unstressed, so leyes might sound like “lays.”

Subcategory IIb. True homophones: In this subcategory, the source of the error is that the phonological form of the phrase is the same for two or more words, and the librarian selects as her interpretation a phrase other than the one intended. The example most familiar to librarians is perhaps euthanasia or youth in Asia. Other well-known examples include China/china, Turkey/turkey, Wales/whales, and Greece/grease. These stories might be dismissed as apocryphal, part of some sort of urban legend in libraries, if it were not for the fact that they were reliably reported in some detail by workshop participants.

In one report, the librarian was thinking about the history and travel sections before the user could say, “I meant china dishes.” In another, there is evidence that the accident is more likely to occur when users use the ambiguous word in a context in which either meaning makes sense: “I want statistics on Wales” makes just as much sense as “I want statistics on whales,” whereas “I want to know about the life cycle of whales” cues the librarian that it is not Wales, since countries do not have life cycles. In the request for statistics, the user averted an imminent communication accident by adding, “I want to know how many whales there are.” A similar cluster of homophones involves animals and sports teams. For example, when the user asked for “Blue Jays,” the librarian dutifully got out the bird books, but should have been looking for material on the baseball team. Tigers, cardinals, orioles and penguins are all fair game for this kind of misunderstanding.

What makes a librarian choose one meaning over another? Like an optical illusion, a homophone presents two meanings from which the viewer or listener almost always perceives one before the other. The interpretation might depend on the librarian’s personal or professional world – blue jays as birds could be more salient than ball teams. Or the interpretation might depend on immediate context: in cognitive science, this phenomenon is explained by the concepts of recency and primacy effects in memory. (Stephen K. Reed, Cognition: Theory and Applications 2d ed. (Pacific Grove, CA: Brooks/Cole Pub. Co., 1987), 85-86.) For example, if the librarian had recently encountered several students working on a nature project where they had to choose a bird as a topic, she might naturally perceive the Blue Jays question to be similar to the bird questions. This explanation also works for the previously quoted example where the librarian mistook “Socrates” for “soccer tees”. In a comparable example, the librarian had been working on a proposal for a new computer system based on UNIX. When the user asked for information on “eunuchs,” the librarian started thinking about UNIX, until the user added “and how they were perceived in Greece and Rome.” A mirror image of this example was collected several years before the UNIX platform was common in libraries. In this case, the user asked, “Do you have information on UNIX?” but the librarian heard “eunuchs.” It follows that, if the alternate term is unfamiliar or unusual, the librarian is likely to choose the more familiar one. The librarian’s choice may also be explained, at least to some extent, by the setting and characteristics of the user, though sometimes there is no apparent reason for one interpretation prevailing.

Some examples of homophone accidents have as their basis lexical or syntactic explanations, as in the one where the user said, “I need books on supervision,” but the librarian heard “super vision” and began thinking about high technology. Since supervision is a familiar word, the explanation that the librarian had never heard this word before is untenable. Likely, this is a case of syntactic misanalysis or possibly a lexical error resulting from where the user placed stress. Supervision is not two words, but it derives from two words, a preposition and a noun, which have been lexicalized into a single word with a special meaning that is distinct from the two separate words. In such cases, the stress often moves from where it would be when the words are separate, which alerts the hearer to the special meaning: compare “white house” and “White House” or “black bird” and “blackbird.” In the case of “super vision” and “supervision”, however, the problem is not initially apparent since there is no stress or pronunciation difference.

Other examples in this category involve alternative meanings of the key verb in the sentence. In one example, the user stated, “I want to cure a fish.” The librarian immediately began thinking about smoking game fish, but the user explained that he wanted to make his pet fish well. In a similar example, the user w anted to “learn to carve game birds,” which could have been either a cooking or woodworking question, depending on which of the two meanings of game birds is intended. This type of communication accident may seem understandable, easily repaired, and even trivial, but in fact it can have negative consequences for the image of the library and the librarian, as well as being an inefficient use of everyone’s time. Consider, for example, the elderly user who obviously had trouble walking, and who asked for the globe. The librarian, assuming she wanted The Globe, a local newspaper, sent the user to another floor, but it turned out that she wanted a globe of the world.

Subcategory IIc. Pseudohomophones. Here the source of the error is that the phonological form of the intended word is very like (but not the same as) that of another word, and the librarian settles on the unintended word. Some accidents in this category are also examples of a local (i.e., arising from a single phrase, word, or letter) input failure, where the listener misses hearing something that is crucial to accurate construction of the intended missing, such as missing a negative like “not”. (In one instance, the user asked, “Do you have a section on house plants?” which the librarian interpreted as a request for “house plans”.)

As in the case for true homophones, we decided what we’re hearing in large part by what would make sense in a particular context, so in the case of pseudo- or almost-homophones, the meaning overrides the minor sound difference, as in the “Socrates”/”soccer tees” example. This often involves a word that is pronounced correctly, but the librarian simply mishears, substituting a similar-sounding word; e.g. insects and incest. “Incest” makes sense to the librarian because people do ask for information on this subject and there is nothing in the context of the request that cues the librarian to the error. As in the category involving true homophones, the librarian may hear “incest” rather than “insects” because she has just had dozens of students asking for information on incest for their high school essays. Again, there is an extralingustic context that makes one interpretation more salient than another.

Other kinds of nonverbal communication cues such as the user’s appearance or age may dispose the librarian to interpret it as being one word rather than the other. This is not surprising since nonverbal messages are an important part of the communication. Research also indicates that, as a general pattern in determining meaning, people usually rely more on nonverbal cues, especially when verbal and nonverbal messages seem to conflict. (Ringle and Bruce, Strategies for Natural Language Processing, 208.)

We do not have sufficient contextual information from the following examples for a detailed explanation, but it is conceivable that the librarian is not reporting (and perhaps not even aware of) the nonverbal signals that affect how he makes sense of the initial query. In a case where the librarian heard the user say “bamboo hearts” and assumed this to be a cooking question, the question was in fact about “baboon hears”. Similarly, we can only speculate about the cues that led the librarian to interpret “herpes” as “hairpiece”. There is, however, another situation that produces this kind of accident: the librarian is not familiar with the word or phrase, and therefore tries to re-form the request into something that makes sense. In one case, the user presents the word carpe diem. The library staff member is apparently unfamiliar with this Latin proverb and thinks it must be somebody’s name; however she cannot find any match in the catalog. It is not clear how this problem was resolved, but it does argue for the need to have staff who are as highly educated as possible working on the reference desk. (Judee K. Burgoon, “Nonverbal Signals,” in Handbook of Interpersonal Communication, ed. L. Knapp and Geralrd R. (Beverley Hills, CA: Sage, 1985). 344-90, 347.) 

III. Secondhand communication accidents
Class III examples are based in errors of interpretation on the part of the user, which are repeated to the librarian. The librarian accurately hears what the user is saying, but the problem arises because the user has previously misinterpreted or failed to remember what he has heard or read, and repeats this misunderstanding. These examples fall into two subcategories.

Subcategory IIIa. Secondhand homophones.
In this category fall examples similar to the Type IIb true homophones like China/china. The difference is that the user hears a word that she interprets incorrectly as being another word that sounds the same but is actually a different word with a different meaning. This phonological identity leads to incorrect meaning selection. Hence, in the instance where the user said, “I was told to ask for Eric,” she seemed vaguely aware that something was not quite right, distrusting what she had heard to the extent that she indicated this to be a “secondhand” question. This can cue the librarian to ask more about the situation; e.g. “Who told you that?” If the answer is, “education prof. … he said Eric would help me find articles,” the accident is easily repaired (unless, of course, Eric is one of the reference librarians).

Subcategory IIIb. Secondhand pseudohomophones.
In these cases, a perhaps unfamiliar phonological form is interpreted by the user to be another form with a more plausible semantic interpretation. Here is where we classify the “oranges and peaches” story, which we will analyze in more detail later. Similarly, it is difficult for the librarian to recognize the communication failure in the case where the user asked for “Tall Stories of Peace and War” because that is a perfectly plausible title. Unfortunately, the user was really looking for Tolstoy’s War and Peace. If he had not rearranged the words but had instead said “Tall Stories of War and Peace,” it might have been easier to recognize, but what had probably happened here was that the user remembered the meaning but not the form of War and Peace. “Tall Stories” represents a syntactic reanalysis in order to make sense of the unfamiliar word “Tolstoy”. Examples in this category appear to result from the user trying to make sense out of what he has heard, when the original words or phrase are unfamiliar. In another example, a young library user asked for information on “the ankle-less horse,” which turned out to be the ankylosaurous, a type of dinosaur. Compared with “oranges and peaches,” this failure is easier to recognize because it seems to make less sense. After all, someone could have written a book called Oranges and Peaches, but there are probably no ankle-less horses.

Category IV. Creative reconstruction
Class IV examples are similar to those of Class IIC insofar as the user “reconstructs” one or more words, or changes the syntax of a phrase. However, unlike Class III examples, their origins does not reside in a “secondhand” communication: the user has not misunderstood what someone else has said. In Class IV examples, the user has originally heard or read correctly, but usually remembers the meaning of the word or phrase, rather than its actual form. The word, phrase, or title is “creatively reconstructed.” What the librarian hears, then, is something that may be correct in meaning but is not expressed in the words that will allow her to match it with the system’s vocabulary.

For example, the librarian has trouble finding the novel Pet Sematary for the user who asks for “Animal Graveyard.” This pair illustrates something important about how memory works: meanings are stored much more effectively in memory than sounds are. That is, people can more easily recall the meaning of something they heard in the past than the actual sounds or physical words that were used to express that meaning. Thus “Animal Graveyard” has (almost) the same meaning as the actual title desire d, Pet Sematary, though it uses completely different vocabulary. In terms of meaning, it should be an easy match between the user and the system, but it isn’t, since systems for matching titles work on the actual words, and not their meanings.

Sometimes, the word substituted “just doesn’t seem right,” so the young user who demanded to know who was “in Clinton’s closet” got a list of U.S. President Bill Clinton’s cabinet, because the librarian readily recognized that the user has used a synonym for the alternate meaning of cabinet. In some examples, the user changes the syntax instead of the words. For example, an academic library user asked, “What is the psychology of makeup?” In fact, the user was misconstruing an unfamiliar technical term, “psychological make-up,” and in trying to impose an interpretation on it, reanalyzed it syntactically into a different grammatical structure. If the user had asked for something like “the psychology of personality,” the librarian might have understood. But instead the user changed the grammar to the extent that the meaning and the form of the original term were thoroughly obscured.

Another example in this category is a child’s request, in a public library, for information on the “Kingdom of Phyla.” In its oral form, it is not immediately obvious that the user is interested in biological classification. Again, this change in syntax obscures the meaning. The last example in this category is a hybrid form in which synonyms, associated words, and other reconstructions are combined to create a real challenge for the librarian: the user asked for the song “The Lady from Esophagus.” Once the librarian thought of the synonym girl for lady, she remembered the song “The Girl from Ipanema.”

An analysis of “Oranges and Peaches”
Let us now return to the “oranges and peaches” story for a more detailed analysis, since it is a classic example of Type IIb failures, which are perhaps most difficult for librarians to recognize and thus to repair. In these cases, the user’s query is ill formed from the system point of view because it is a communication accident carried over from a previous situation. In this previous situation, the user failed to hear what was said by someone else. The instructor did say “On the Origin of Species,” but the listener (who becomes the library user) hears “Oranges and Peaches,” which he dutifully copies down, perhaps with some puzzlement. Even though the title seems somewhat odd to the user, it is plausible that someone called “Charles something” might have written a book called Oranges and Peaches that is “the Bible of evolution”.

It seems reasonable to the user, therefore, to ask for Oranges and Peaches, especially since the instructor has suggested that the library will have “hundreds of copies”. Interestingly, it also seems plausible to the librarian that a book might be titled Oranges and Peaches and that the library might hold this book. The library, however, has no such record. In the librarian’s mind arise two main possibilities: either Oranges and Peaches is the correct title and the library does not hold the book, or the user is mistaken in the title. From previous experience, the librarian knows that users often make mistakes in the title, provide misquotes, or give dates that are considerably off the mark. (This observation has implications for the recent adoption by many of libraries of a “two-tiered” reference service, where nonprofessional or paraprofessional staff handle the front lines, referring only “difficult” reference questions to professional librarians who stay behind the scenes. A major fallacy of this approach, in our view, is the assumption that difficult questions are easily recognized as such.) But without further context, the librarian has no way of guessing what the correct title might be.

In the user’s mind, there is only one possibility: the librarian is unable to find the book for some reason, probably because he is incompetent. The user’s frustration is increased by the urgency involved; instead of accepting the librarian’s statement that the library has no such book, he begins to berate the librarian, and in so doing, accidentally reveals sufficient information for the librarian to deduce the real title and author. An interesting point here is that the librarian is able to translate Oranges and Peaches into On the Origins of the Species, which he knows, through his own education, is “the Bible of evolution.” This deduction is confirmed by the additional information that it was written by someone named Charles. Once the connection is made, the librarian is able to link the user with the book. (Note again the value of a good general education in solving problems at the reference desk!)

The mental processes here are far more complicated than we have described but this brief analysis suggests how human beings, through talking with each other and picking up nonverbal cues, are able to repair communication accidents. Part of this ability comes from experience. Because librarians deal with so many questions from so many users of differing cultural backgrounds, educational levels, and varying states of mind, they become constantly alert to very subtle cues that the initial question is not well formed.

But what are the consequences if the accident is not quickly repaired, or not repaired at all? In the “oranges and peaches” story, the user became agitated and persisted, but he might equally have turned away, believing the library did not have the book he needed. Clearly, if the librarian is unable to figure out what the user had originally been told, the user does not get what she or he needs from the library. It is therefore important to be able to explain this type of accident, why it happens, what could be done to prevent it from happening, or what could be done to maximize the chances of a successful outcome. Frequent negative outcomes create systemic problems: the reputation of the library for poor service (in the mind of the user, and perhaps in the minds of other users who overhear the exchange) together with the reputation of library users as “stupid” (in the mind of the librarian) form a feedback loop which in fact may create the reality of poor service.

Strategies for dealing with type I and type II accidents
In the following section, we comment on circumstances under which communication accidents are likely to occur, and suggest some specific strategies for averting or repairing communication accidents of Types I and II. Strategies for handling Type III examples (secondhand accidents), and IV (creative reconstruction) are presented in the next section.

Instead of thinking of the circumstances under which accidents are likely to occur in terms of the personal characteristics of the user (e.g., blaming the user for being uneducated, inarticulate, inattentive, or lazy), we suggest that it is more helpful to look at the rules of governing general conversations and how those are amended in the context of the reference transaction.

It is interesting to note that many of the examples of Type II failures (e.g. Wales/whales, house plans/house plants), the form of the initial query leaves it vulnerable to misunderstanding. Users typically present the question as a subject inquiry or directional request devoid of further context that would enable the librarian to infer the intended meaning. The user who wanted house plant information did not say, “I have whiteflies on my aspidistra. Where are your books on house plants?” Instead, the formula for initial requests is commonly (1 and 2) or (3):

1. a. Do you have…?
b. Where is…?
AND
2. a. information on X
b. a book on X
c. a section on X
OR
3. a. I want to find out about X
b. I need information about X
c. I need/want a book about X

William A Katz, “Reference Services and Reference Processes,” Introduction to Reference Work, vol. II, 5th ed. (New York: McGraw-Hill 1987), 51. 

These variants are sometimes classified as holdings, directional, or substantive requests. However, another way of thinking about the problem is to observe that there are two basic and overlapping explanations for the fact that users construct their initial questions in these forms, without providing any further contextual clues to the meaning of the subject X. First, as the linguistic Thomas Eichman has pointed out, the function of an initial question in a reference encounter is not primarily to ask the question; rather, it is a form of phatic communication. (Thomas Lee Eichman, “The Complex Nature of Opening Reference Questions,” RQ 17 (Spring 1978): 217.) 

That is, sometimes people speak just to achieve some social goal, such as (in this case) establishing contact with the librarian. For this purpose, a long (or detailed) description of what is wanted is not necessary. In an attempt to help practitioners cope with this behavior, Ross and Dewdney have worked out an explanation and remedy for this phenomenon (Catherine Sheldrick Ross and Patricia Dewdney, Communicating Professionally: A How-To-Do-It for Librarians (New York: Neal-Schuman, 1989.). Their recommendation is that the librarian treat the initial question not as an accurate representation of the information need, but as four unspoken questions being asked by the user:

Am I in the right place?

Are you available to help me?

Have we made contact? (Are you listening and willing to help me?)

Have you understood my topic (in general)? I’m going to be describing a problem about X.

One effective response is represented by a mnemonic, PACT. The librarian must respond in a way that indicates the user is in fact in the right Place, the librarian is Available to help, the user has made Contact in that the librarian is listening and willing to help, and the librarian has heard the Topic, or at least has some idea of what the query is about. (Ibid, 109-110. In using the PACT technique, the librarian also has the opportunity to indicate if she is not the right person to be asking, or not currently available, and to provide alternative directions (e.g. “I am on the phone right now but will be available in a minute if you can wait.”))

The librarian must not only recognize the need to answer these unspoken questions, but must quickly and clearly show to the user that she has recognized this need. A simple formula for doing this is to respond by repeating some of the user’s words, along with the appropriate body language that invites the user to describe the information need in further detail, (e.g. “Yes, we have quite a large section on house plans.”) The repetition of some of the user’s initial words is a standard interviewing technique called restatement or acknowledgement (Allen E. Ivey, International Interviewing and Counseling: Facilitating Client Development, 2d ed. (Pacific Grove, CA: Brooks/Cole, 1988): 93-94; and Ross and Dewdney, Communicating Professionally: A How-To-Do-It for Librarians, 36-37). The general purpose of this type of response is to establish a good communication climate, but specifically the restatement allows the user either to confirm the librarian’s understanding of the request, or to correct it (“Not house plans – house plants!”) before time is wasted on developing a search strategy, looking in the catalog, or giving directions. The librarian can at the same time appropriately respond to a common social function of the initial utterance (the need to establish contact), while at the same time furthering the query negotiation function of the reference interview.

The second and complementary explanation for the fact that users frequently begin their reference questions with “Do you have…,” “Where is…,” or “I need information on…” is related to the mental model that users have of libraries and librarians. We are currently investigating these mental models as represented by the user’s side of the reference interview, but here we provide a preliminary explanation of how they work. Our hypothesis is that users believe that a library’s resources are for the most part organized by subject, but sometimes organized by format, and that separate physical locations correspond to different subjects or formats. (This is, of course, a fairly accurate but simplistic model that fails to accommodate the complexity of library organization.) Hence users ask questions about where they can find a particular subject (e.g., whales) or format (e.g. newspapers). Underlying such questions are several assumptions, such as that information is found in books, which are found in libraries, and so on. But two more important assumptions made by the user are

1. that if only he can locate the section or subject area, he will be able to find the answer to the more specific question by himself, and
2. that librarians (or anyone who appears to work in a library) know where these locations are because that is their job.

This rather one-dimensional view that the user holds of the librarian’s job and the nature of libraries has been supported by previous research (See, for example, Roma. Harris and Christina Sue-Chan, “Cataloguing and Reference, Circulation and Shelving: Public Library Users and University Students’ Perceptions of Librarians,” Library and Information Science Research 10 (Jan.-. 1988): 95-107.) Consequently, the user reasons that if he states the subject, usually either very generally (e.g. “Where is the law section?”) or very specifically (“I’d like to see the World Almanac”), the librarian will be able to identify the physical location, and the user will then be able to answer the question without further assistance. In some cases, the reasoning is that if a particular reference tool helped answer a previous question, it will also answer the current question. From the librarian’s point of view, of course, the organization of sources is much more complex, and more options are available than the user realizes. It is not the purpose of this paper to examine the truth or untruth of these beliefs, but it is important to the understanding of communication accidents to recognize that users, by and large, present their information needs as questions about subjects or locations rather than as problem-centered statements.

Problem-centered statements (e.g. “I want to get rid of the whiteflies on my aphidistra”) give the librarian more context for the query and therefore provide a greater chance of a successful outcome, and less likelihood of a communication accident (For an explanation of how the librarian can enhance chances of a successful search by inviting the user to describe the situation underlying the reference question and the uses to which the user intends to put the information sought, see Brenda Dervin and Patricia Dewdney, “Neutral Questioning: A New Approach to the Reference Interview,” RQ 25 (Summer 1986): 506-13). The pragmatic explanation of an initial query as an attempt to establish contact and the mental-models explanation are not inconsistence. In either case, when a Type II accident occurs and is unrecognized by the librarian, she can select one of at least three alternative responses to the initial request, as she has perceived it. For example, if the user says X and the librarian hears Y, the librarian may:

1. respond literally to what she has heard without restating or asking questions (e.g. “Yes, we have that/ Here is the class number [for Y].”
2. start talking about Y instead of asking more about X. (“Oh, are you going on a trip to Wales?”) This sometimes provides the user with a chance to correct the librarian.
3. ask more about Y, thereby inviting the user to talk about X. This almost always elicits context from which she can distinguish between X and Y (e.g. “What kind of statistics about Wales would you like?” “I’d like to know how many whales there are,” or even “Everything you’ve got – I have to do a biology project.”)

The librarian’s best bet is to choose 3: the reference interview. The context then emerges, even if the librarian is unaware of the ambiguity or the mishearing. Restating the user’s initial words, in combination with an open question that encourages the user to talk, is almost invariably a fail-safe behavior. The second best approach is to have some conversation with the user, in which it may emerge that the topic is X, not Y. The least effective approach is to not talk about Y at all, thus providing no opportunity to repair the misunderstanding. Whichever option is chosen by the librarian, she should at least conclude the transaction with a follow-up question, which may be the single most important interview skill because it allows the librarian, belatedly if necessary, to repair the misunderstanding (Ralph Gers and Lillie J. Seward, “Improving Reference Performance: Results of a Stateside Study,” Library Journal 110 (Nov. 1, 1985): 34). Some recommended follow-up questions are:

Does this completely answer your question?

Does that help you?

Will that be all you need? (Various forms of follow-up questions are suggested by Gers and Seward, 34, “Improving Reference Performance”, Laura J. Isenstein, “Get Your Reference Staff on the STAR track,” Library Journal (Apr. 15, 1992): 36; and Ross and Dewdney, “Communicating Professionally,” 113.)

In summary, the chance of a communication accident is decreased, or at least the negative consequences of that accident is minimized, if the librarian acknowledges the user through restatement, uses open questions to find out more about what the user wants to know, and asks follow-up questions to ensure that the user finds what he is looking for.

Strategies for dealing with type III and IV accidents
Communication accidents of Type III (secondhand accidents) and Type IV (creative reconstruction) present a more difficult situation. Although strategies such as asking open questions are often useful for Type III and IV accidents, the user may be unable to respond to the librarian’s questions in any knowledgeable way because he simply does not know the answers to the questions. For some reason, the user has “got it wrong”. Obviously, if the librarian restates “Oranges and Peaches,” the user will affirm this title instead of correcting it, because that is what he heard. This may also happen with initial questions that seem to make less sense. In a rather unusual one, the user asks, “What does ‘pirandello’ mean?” A knowledgable librarian can restate “Pirandello?” or even add, “That’s the name of a playwright,” but this does not help the user who had read, in Arsenic and Old Lace, that “so-and-so was felling very ‘pirandello,’” nor would restatement help the librarian who shared the user’s assumption that this was a standard adjective.

But in both of these examples, certain kinds of open questions work well. People always ask reference questions from a personally meaningful context, and the context can be uncovered by using neutral questions, described by Dervin as open questions designed to get at three dimensions of the underlying information need: the background situation (how the information need was generated), the gap or missing piece of understanding, and the uses, or how the user plans to use this information (Dervin and Dewdney, “Neutral Questioning: A New Approach to the Reference Interview). A neutral question that almost always works in a Type III (secondhand) situation is, “How did you hear about/Where did you read about X?”Knowing that the information need is embedded in a particular situation helps the librarian to detect the associations – “oranges and peaches” have to do with evolution, and Pirandello was used in a sentence as an adjective to evoke a feeling.

Another strategy commonly used by librarians with students is to ask what the instructor actually said, or to ask to see what is written in the assignment. An excellent example of efficient context appears in a case where a library technician was asked for information on “Dart’s Child.” The open question, “What is Dart’s Child?” elicited the response, “I don’t know. If I did, I wouldn’t be asking you, would I?” The technician then tried, “Where did you hear this phrase?” and was told that it was mentioned in an anthropology lecture. Given this context, she used a social sciences encyclopedia to discover excellent information on Dart’s Child (more commonly known as the Taung baby) under an entry about Raymond Dart, the paleoanthropologist.

Conclusion
Librarians sometimes question the appropriateness of telling stories about communication accidents, many of which are humorous. In a recent exchange on LIBREF-L, where a subscriber raised the ethical quest ion of whether librarians ought to be laughing at library users repeating funny stories, some librarians replied, “Lighten up – doctors have their patient stories, mechanics have their car owner stories, and we all need to let off steam.” We agree. Sharing such stories is in fact useful, not only to “let off steam,” but also to provide concrete examples of communication accidents for analysis so that librarians recognize what is really going on, why it happens, and what would constitute a helpful and professional response. Stories like “oranges and peaches” are amusing, but laughing at the discrepancy between “Oranges and Peaches” and Origin of the Species is not the same as ridiculing the user. It is not useful (let alone professional) to blame users who are making sense as best they can out of a situation which they may not fully understand. All of us, from time to time, find ourselves in the position of users dealing with systems whose functions and arrangement are not completely clear to us. The stories that result from these communication accidents are instructive, and analysis of these stories, such as typology that we have suggested in this article, may be helpful in making us better human service professionals.

About the authorsAt the time the article was published, Patricia Dewdney was an associate professor at the Graduate School of Library and Information Science and a faculty member in Communications and Open Learning at the University of Western Ontario in London, Ontario. This was the second time she received a Reference Service Press Award; she is also a co-recipient of the 1996 award.

Gillian Michell was an associate professor and the chair of the graduate program in Library and Information Science at the University of Western Ontario.

The authors wish to thank the Social Science and Humanities Research Council of Canada for their financial support of this research, which was part of a larger project, “Mental Models of Information Systems: A Study of Ill-formed Queries Presented in Libraries,” #413-93-0088.