orality-hcid-itid09

37

Jahanzeb [email protected] FacultyLanguage TechnologiesInstituteCarnegie Mellon University5000 Forbes Ave.Pittsburgh, PA 15213 USA(412) 567-1204

Nosheen [email protected] ResearcherDepartment of DevelopmentSociologyCornell University119 Warren HallIthaca, NY 14853 USA(650) 996-7122

Carolyn Penstein Rosé[email protected] ProfessorLanguage TechnologiesInstitute and Human-Computer Interaction InstituteCarnegie Mellon University5000 Forbes Ave.Pittsburgh, PA 15213 USA(412) 268-7130

Roni [email protected] of Computer ScienceCarnegie Mellon University5000 Forbes Ave.Pittsburgh, PA 15213 USA(412) 414-3701

Orality-Grounded HCID: Understanding the Oral User SHERWANI, ALI, ROSÉ, ROSENFELD

Research Article

Orality-Grounded HCID:Understanding the Oral UserAbstract

While human-computer interaction (HCI) methodologies are designed to begeneral, they have most often been applied in the context of literate end usersin the West. These methodologies may, however, need rethinking for applica-tion in HCI for the developing world (HCID) contexts, where many of the basicassumptions that underpin the methods may not always hold true. In this arti-cle, we present an overview of one factor that is signiªcantly different in theHCID context—the literacy of the end user—by drawing on the literature oforality, and we offer a framework for HCID methodology that we argue ismore appropriate for the HCID context. Based on this framework, we thenpresent guidelines for design and user research methodologies in such con-texts, highlighting seminal HCID research that corroborates these guidelines.

IntroductionOne of the most valuable contributions from the ªeld of human-computerinteraction (HCI) to date is the collection of methodologies for user-centered design that have evolved through generations of design work thatfocus on the needs of users within their context of work or life. While, atthe most abstract level, these methodologies have been designed to begeneral, they have most frequently been applied to populations within thedeveloped world. The danger in such a population bias is that it may leadto certain blind spots that can result in naïve applications of these practicesin the developing world if practitioners do not deeply consider how thepractices must be parameterized in order to generalize and be effectivewith these very different populations. The contribution of this article is toraise awareness to these blind spots by contrasting common assumptionsunderlying HCI as it is practiced within the developed world with recentªndings from work in HCI for the developing world (HCID).

In HCID literature, users unfamiliar with the tools of literacy—readingand writing—are commonly described as illiterate, non-literate, low-literate, or semi-literate, and it is these users that most HCID interventionsfocus on. We suggest that a better understanding of these users can beachieved through the concept of orality, which has been notably theorizedby Walter Ong in his seminal work Orality and Literacy (1982). In this arti-cle, we use this framework as a lens with which to re-examine commonpractices from mainstream HCI.

Orality describes how people think, communicate, and learn in a cul-ture where writing has not become internalized. Orality theory arguesthat writing has so fundamentally transformed consciousness in literatecultures that we—literate researchers—are unable to grasp how oral cul-tures and people operate. The very categories through which such peopleare described is a case in point: They are described in terms of what they

© 2009 USC Annenberg School for Communication & Journalism. Published under Creative Commons Attribution-NonCommercial-Share Alike 3.0 Unported

license. All rights not granted thereunder to the public are reserved to the publisher and may not be exercised without its express written permission.

Volume 5, Number 4, Winter 2009, 37–49

are not, such as “illiterate,” rather than what theyare, that is, oral. Terms like “illiterate” devalue theidentity and knowledge of oral cultures by implicitlysuggesting that lack of literacy is equivalent to back-wardness. Orality theory emphasizes that we needto understand such communities in their own terms,rather than from the perspective and biases of liter-ate users.

Ong discussed orality in the context of oral cul-tures, how they contrast with literate cultures, andhow modern media is creating a “secondary orality”for literate cultures. While many HCI researchershave referenced Ong’s work, the focus has been onthe implications of orality for literate societies, espe-cially on the notion of secondary orality in such con-texts, and there has been little exploration of Ong’swork with respect to oral cultures. In this article, weattempt to begin this exploration by leveragingOng’s concept of orality as a foundation on whichto build a model for oral users in HCID contexts,and to use this model to develop recommendationsand predictions for user interface design, as well asuser-study methods in these contexts.

Related WorkOng (1982) presented an extensive contrastbetween what he referred to as oral cultures andwhat he referred to as literate cultures, a contraston which we base much of our proposal andthrough which we interpret ªndings from our andother people’s ªeldwork in the developing world.Ong himself did not mean his synthesis primarily tobe used in the context of work in the developingworld, but rather painted a picture of how modernmedia was creating a secondary orality for literatecultures. In that vein, Ong’s work has already playedan inºuential role within the ªeld of HCI, facilitatingthe application of insights from oral cultures todesign within the developed world, increasing sorecently in the light of conversational Web 2.0 tech-nologies. Sometimes Ong’s insights have beenapplied at a general level (e.g., Mathur &Karaholios, 2009; Conrad, 2003), simply taking asingle quote from his work and using it as inspira-tion for an otherwise bottom-up, user-centereddesign process. Several researchers in the past dec-ade have discussed the emergence of Web 2.0 tech-nologies (e.g., Stahl, 1999; Fischer, 2006; Wright,2008; Bergstrom & Karahalios, 2009), pointing tothe conversational natural of those media, and

framing those media within the vision for a second-ary orality. Other times, his framework has beenadopted or scrutinized at a more detailed level (e.g.,Davis, 1997; Grudin, 2007; Wright, 2008), some-times challenging or extending Ong’s ideas. In muchof this work, Ong’s vision has not been adoptedwholesale, but rather has been used as a catalyst fordiscussion, often about the future of technology. Forexample, Grudin (2007) added a third distinction toOng’s two-way distinction between oral cultures andliterate, or digital, cultures, and used this to makepredictions about trends in computer usage by mod-ern youth. Ackermann & Decortis (2007) similarlyexplored Ong’s ideas in connection with e-literacyand youth, looking at opportunities for youth toengage in storytelling behaviors on the Web. Davis(1997) reinterpreted some of Ong’s ideas in terms ofwhat they might mean for video as a form of com-munication. In contrast to this body of work, whichtook insights from the developing world and usedthem to provide new insights into design for thedeveloped world, our work brings these insights fullcircle, applying them within their context of origin,the oral cultures themselves.

One of the initial efforts to create design recom-mendations for illiterate users drew primarily fromliterature on screen-readers for the visually impairedand from sighted users working without visual dis-plays, since “human computer interaction and user-interface design literature provided little speciªcresources for designers interested in building soft-ware for illiterate users or members of the develop-ing world” (Huenerfauth, 2002). However, bothvisually impaired users in the developed world andusers working without visual displays are literategroups of users. Furthermore, these strategies werenever tested on actual oral end users.

Within the ªeld of evidence-based HCID research,there has been much work on the speciªcs of indi-vidual projects and artifacts, both in terms of inter-face design and evaluation, such as HealthLine(Sherwani et al., 2007; Sherwani et al., 2009) andDigital Green (Gandhi et al., 2007). There is also asubstantial body of work that goes beyond the spe-ciªcs of one artifact, and aims to provide generalizedguidelines for interface design and evaluationgrounded on empirical, ªeld-based research, includ-ing the following: design guidelines for text-freeinterfaces (Medhi et al., 2006), user interface (UI)strategies for low-literacy users (Parikh et al., 2005),full-context video (Medhi & Toyama, 2007), and the

38 Information Technologies and International Development

ORALITY-GROUNDED HCID: UNDERSTANDING THE ORAL USER

Bollywood and auto-rick radio measurement methods(Chavan, 2007). While these research contributionsdo not provide a theoretical framework that explainswhy these methods work well, they do provide valu-able insights that can be used to build such a theo-retical framework. We will now summarize some ofthe ªndings from these projects to provide contextfor the discussion in the remainder of this article.

Digital GreenIn the Digital Green project (Gandhi et al., 2007) foragricultural extension, researchers at MicrosoftResearch in India designed, developed, and evaluateda mechanism for generating and disseminating video-based instructional content for teaching agriculturalbest practices to farmers. While watching a video didnot always lead to behavioral change on its own, theresearchers’ “ªnding was that mediation [through alocal facilitator] is essential to the process of exten-sion that farmers were most convinced by appropri-ately targeted and pitched content.” Also, “villagemediators encourage these farmers to share their per-sonal experiences to motivate their peer groups.”Finally, during video screenings, “viewers frequentlyask for the names and villages of recorded farmers.”

Full-Context VideoThe “full-context video” approach to providing tuto-rials for an interface (Medhi & Toyama, 2007) hasbeen shown to lead to remarkably improved tasksuccess rates; one out of 17 participants was able tocomplete the job-ªnding task without the help of afull-context video, while 18 out of 18 participantswho were shown the video were able to completethe same task—an improvement from 6% task suc-cess to 100%. In that study, the full-context videoshowed not only the mechanics of how to completethe job-search task, but more importantly, also pre-sented a dramatization of the context of the task:An employer uses an interface to post a job, a job-seeker is told by a friend how to use the interface toªnd the job, and the story ends with the job-seekerªnding the employer and getting the job. Medhiand Toyama concluded:

Overall, the video appeared to instill a greatamount of conªdence among the test partici-pants. Subjects were seen to gesture in agree-ment, to smile, and to laugh at various points inthe video. It was as if the video provided a shift inconcern, from anxiety about how to use the de-vice to concerns about the content itself. . . . Sub-jects who saw the video ªrst appeared more

determined—eyebrows knit, body leaning intothe monitor—when moving onto the applicationitself, as compared with the control group . . .More than one subject explained that they hadnever understood what computers were for, butthat the video showed why people use them. Thedramatized video made this clear (p. 2).

HealthLineIn our own work (Sherwani et al., 2009), we havebeen designing, developing, and evaluating speech-based health information access interfaces for low-literate community health workers. Our research hasbeen conducted in rural parts of the Sindh provincein Pakistan, where we have conducted user studieswith up to 20 participants in each, all of them com-munity health workers. These community healthworkers are often not able to read at all, or do sowith great difªculty.

We have seen how human-guided tutorials—where the participant uses the interface with theactive guidance of the facilitator, as a trainingtool—outperform verbal instruction, written instruc-tion, and even video-based instruction (the video wetested was not full-context video). Further, we haveseen how oral participants have greater difªcultycomprehending the same material even when theyspeak the language as ºuently as their literate coun-terparts, and how content designed for literate con-sumption needs to be signiªcantly adapted to beused by oral users. It was our experience in thesestudies, and the difªculty in applying the tools usedin standard research and design practice, that led ustoward the work on orality.

We now turn to a discussion of orality, to presenta framework for oral users, after which we will pres-ent our recommendations and predictions, in thecontext of existing work.

Orality versus LiteracyIn this paper, we argue that HCID research can besubstantially improved by grounding it in an under-standing of orality. While there is a continuumbetween completely oral users and completely liter-ate ones, in this paper we will use the term “oraluser” to refer to aspects of completely oral usersthat are present even in somewhat literate people.

To design for oral users, we must ªrst understandtheir cultural practices of community knowledge-building and transmission, including how they(a) organize and transmit information, (b) learn infor-

Volume 5, Number 4, Winter 2009 39

SHERWANI, ALI, ROSÉ, ROSENFELD

mation, and (c) remember information. Additionally,we need to grasp what Ong (1982) referred to as the“psychodynamics of oral thought” to understand thefundamental differences between oral and literateusers as individuals within that cultural context.Below, we expand on each of these aspects.

Cultural Practices of Knowledge Buildingand Transmission

1. How information is organized andtransmitted

A key difference between oral and literate cul-tures is how knowledge is organized and trans-mitted. The sum of an oral community’sknowledge—its history, its identity, its culture,and its whole way of life—is represented andremembered in the oral tradition, often referredto as a culture’s folklore. Among other dimen-sions, this oral tradition comprises stories, myths,proverbs, riddles, poems, and songs. These formsprovide a pool of wisdom that individuals candraw on as a means of daily survival, as well asdeal with new situations.

Notably, this knowledge is not staticallystored; rather, it is kept alive through continualverbal, extempore performance. Storytellersdynamically render different parts of this knowl-edge at different occasions, altering the contentbased on the current political scenario, socialsentiment, and immediate audience. Since thisknowledge is never written down, it is ªxed nei-ther in form or content, nor is it repeated wordfor word. Hence, knowledge is dynamic andevolving, ºuid and creative, and it is composediteratively over generations.

One of the striking characteristics of oralknowledge is that it “knows no lists or charts orªgures” (Ong, p. 97). The emphasis is on narrat-ing events in time, with a correspondencebetween these narratives and human experience.

2. How information is learned

According to Ong (1982), “human beings in pri-mary oral cultures . . . learn by apprenticeship . . .by discipleship, by listening, by repeating whatthey hear, by mastering proverbs and ways ofcombining and recombining them, by assimilatingother formulary materials, by participation in akind of corporate introspection—not by study inthe strict sense.” What is especially relevant toHCID practitioners is what is not mentioned in the

above quotation: Oral people do not learn from aneutral, stand-alone object, such as a book, orautomated system, which contains a set ofabstract instructions to be applied across situa-tions; rather, they learn “in situ,” embedded inconcrete situations and practical experience.

3. How information is remembered

Since information in oral cultures is transmittedthrough oral delivery—and hence it is recitedand heard—it needs to be presented in a formthat is conducive for aural-oral reception andretention. Ong elaborated:

In a primarily oral culture, to solve effectivelythe problem of retaining and retrieving care-fully articulated thought, you have to do yourthinking in mnemonic patterns, shaped forready oral recurrence. Your thought mustcome into being in heavily rhythmic, balancedpatterns, in repetitions or antitheses, in allit-erations or assonances, in epithetic and otherformulary expressions, in standard thematicsettings . . . in proverbs which are constantlyheard by everyone so that they come to mindreadily and which themselves are patternedfor retention and ready recall, or in othermnemonic form. Serious thought is inter-twined with memory systems. Mnemonicneeds determine even syntax. (Ong, p. 34)

In short, the usage of rhythm, repetition, locallyknown idioms, and dramatized settings are keyto keeping orally constructed knowledge aliveand remembered. Even though these mecha-nisms can aid memory in literate contexts, theability to store information externally throughthe technology of writing means that suchmechanisms are not as crucial in literate contextsas they are in oral contexts.

Psychodynamics of Oral Thought(Contrasted with Literate Thought)

1. Additive, not subordinative

While the preferred grammatical form of literatewriting is subordinative, oral thought prefers addi-tive forms. Ong’s example from an older transla-tion of the Bible—one considered to have a largeamount of oral residue—contrasts neatly with amodern translation. The older translation reads:

In the beginning, God created heaven andearth. And the earth was void and empty,



and darkness was upon the face of the deep;and the spirit of God moved over the waters.And God said: Be light made. And light wasmade. And God saw the light that it wasgood; and . . . (Ong, p. 37)

The word “and” is used to join phrases and sen-tences liberally, which is not the case with amodern translation:

In the beginning, when God created theheavens and the earth, the earth was a form-less wasteland, and darkness covered theabyss, while a mighty wind swept over thewaters. Then God said, “Let there be light”,and there was light. God then separated thelight from the darkness . . . (Ong, p. 37)

Here, the use of conjunctions—when, while,then—to join phrases is the common case, witha more hierarchical and subordinative organiza-tion than the additive pattern shown in the pre-vious passage. Ong argues that the formerexample is as natural to the oral mind as the lat-ter example is to the literate mind.

2. Aggregative, not Analytic

Oral tradition favors the use of reusable formu-las. Thus, it is not just “the soldier,” but “thebrave soldier”; not just “the princess,” but “thebeautiful princess”; and not just “the oak,” but“the sturdy oak.” These clusters aid memory bycreating archetypal concepts that are repeatedlyused in narratives. Furthermore, according toOng:

Traditional expressions in oral cultures mustnot be dismantled: it has been hard workgetting them together over the generations,and there is nowhere outside the mind tostore them. . . . Once a formulary expressionhas been crystallized, it had best be kept in-tact. Without a writing system, breaking upthought—that is, analysis—is a high risk pro-cedure. (p. 39)

These reusable formulae are, in some ways, thefundamental building blocks of oral tradition.

3. Redundancy

All spoken information is ephemeral, and ifinformation were never repeated, even theslightest loss of concentration would mean acomplete misunderstanding on the part of thelistener. Redundancy is necessary to ensure thatthe speaker and listener both remain on track.When reading, however, it is not necessary torepeat information, as the reader can alwaysbacktrack to re-read earlier material—to thepoint where this becomes second nature. Thediffering economies of hearing and readingresult in a signiªcant difference between theoutputs of oral and literary content in terms ofthe redundancy inherent in the content. Eventhough redundancy may be a natural part of thethought process, “with writing, the mind isforced into a slowed-down pattern that affords itthe opportunity to interfere with and reorganizeits more normal, redundant processes” (ibid.,p. 40).1

4. Conservative/Traditionalist

A corollary of redundancy is conservatism. Sinceoral knowledge vanishes unless repeated againand again, oral cultures place a premium onrepeating previously held knowledge rather thanexperimenting and discovering new knowledge.To the technologically minded, this can best beexplained with an analogy to hard drive capaci-ties. When hard drive space was expensive, onehad to keep only a limited amount of data,which was most likely static; however, oncecapacity became cheap, it was easier to keepstoring newer and newer data, as there was noneed to choose between the old and the new.Oral cultures function with the former mindset.

This is not to say that oral cultures neverupdate their knowledge store or that they lackoriginality:

Narrative originality lodges not in making upnew stories but in managing a particular in-teraction with this audience at this time—atevery telling the story has to be introduceduniquely into a unique situation, for in oral



1. The content used by NGOs for information dissemination is often in a highly summarized format in an attempt tocrystallize the salient points that need to be conveyed. However, this is not the optimal format to present informationto oral users. This has been our experience in the HealthLine project, and researchers at the Meraka Institute in SouthAfrica faced the same issue in the OpenPhone HIV Information Line project (Aditi Sharma Grover, personal communica-tion).

cultures an audience must be brought to re-spond, often vigorously. But narrators also in-troduce new elements into old stories.” (ibid.,p. 41)

5. Close to the Human Lifeworld

In the absence of elaborate analytic catego-ries that depend on writing to structureknowledge at a distance from lived experi-ence, oral cultures must conceptualize andverbalize all their knowledge with more orless close reference to the human lifeworld,assimilating the alien, objective world to themore immediate, familiar interaction of hu-man beings. . . . An oral culture has no vehi-cle as neutral as a list. . . . [It] likewise hasnothing corresponding to how-to manuals forthe trades. (ibid., p. 42)

Trades are learned primarily through apprentice-ship, involving observation and practice, withminimal verbalized knowledge, if any. Further-more, oral cultures do not preserve knowledgeof skills in abstract, self-contained corpora.Finally, oral people do not treat all humanlifeworld-based information as equally impor-tant; it is their speciªc lifeworld that is relevant,and hence, important. For instance, they givespecialized names to details that matter in theirspeciªc lives, but generalize others that do not(e.g., “that is merely a ºying animal”).

6. Agonistically Toned

To literate individuals, oral people might appearto be extraordinarily agonistic or argumentative.This is because stories and proverbs—the basisof oral tradition—are not meant to be just astore of knowledge, but also a means to engagea dialectic dialog. It is this argumentation thatOng considers to be the predecessor of the dia-lectic method of Socrates and Plato. Thus, oralthought is often instantiated as an interplay ofcompeting ideas.

7. Empathetic and Participatory

With literacy, knowledge is disengaged and issupposedly “objective.” In an oral culture,according to Ong, “learning means achievingclose, empathetic, communal identiªcation withthe known.” Oral people usually do not memo-rize, other than for rituals; instead, learninginvolves an amalgamation of the new with theself.

Literates are usually surprised to learn thatthe bard planning to retell the story he hasheard only once wants often to wait a day orso after he has heard the story before hehimself repeats it. In memorizing a writtentext, postponing its recitation generally weak-ens recall. An oral poet is not working withtexts or in a textual framework. He needstime to let the story sink into his own store ofthemes and formulas, time to ‘get with’ thestory. In recalling and retelling the story, hehas not in any sense ‘memorized’ its metricalrendition from the version of the other singer.(ibid., p. 59)

8. Homeostatic

On the order of generations, oral cultures elimi-nate memories that are obsolete, simply by notrepeating that information. As a consequence,oral cultures never need dictionaries, since allwords in usage are commonly known and under-stood by everyone in the culture. The number ofwords never grows unmanageably, as obsoletewords are pruned away. Literate cultures, on theother hand, can access all words from the presentday back thousands of years ago, and require dic-tionaries to store and access these words.

9. Situational, not Abstract

Perhaps the most fundamental differencebetween oral and literate cultures is that of situ-ational thinking versus abstract thinking. Becauseoral knowledge is rooted in the human lifeworld,oral people are most comfortable in thinking andlearning situationally, instead of in the abstract.This does not mean that oral people cannotthink categorically; it is just that they make cate-gories differently than literate people. For exam-ple, when asked to categorize a set of objects,oral people categorize based on features impor-tant to the human lifeworld (e.g., usefulness) asopposed to abstract features (e.g., function). Forinstance, Ong wrote that:

Subjects were presented with drawings offour objects, three belonging to one categoryand the fourth to another, and were asked togroup together those that were similar andcould be placed in one group or designatedby one word. One series consisted of draw-ings of the objects hammer, saw, log,hatchet. Illiterate subjects consistently





thought of the group not in categoricalterms (three tools, the log not a tool) but interms of practical situations—“situationalthinking”—without adverting at all to theclassiªcation “tool” as applying to all but thelog. If you are a workman with tools and seea log, you think of applying the tool to it, notof keeping the tool away from what it wasmade for—in some weird intellectual game.(Luria, 1976 in Ong, 1982)

Further, any form of thought other thangrounded “operational thinking” is likely to beconsidered “not important, uninteresting,trivializing.” (Ong, 1982)

Operational thinking is also the mode inwhich oral individuals interpret reality. For exam-ple, Luria found that oral people interpreted cir-cles as a plate, sieve, bucket, watch, or moon,although school-going children readily identiªedthem as circles (Luria, 1976).

Likewise, oral people do not think in formal,syllogistic logic. For instance, when given thequery, “precious metals do not rust, gold is aprecious metal—does it rust or not?” one oralrespondent said: “Do precious metals rust ornot? Does gold rust or not?” Another said, “Pre-cious metal rusts. Precious gold rusts.” Syllo-gisms are special riddles where the conclusionsare derived from the given premises alone.Ong said:

Persons not academically educated are notacquainted with this special ground rule buttend rather in their interpretation of givenstatements, in a syllogism as elsewhere, to gobeyond the statements, as one normally doesin real life or in riddles (common in all oralcultures). (p. 53)

Furthermore, requests for deªnitions are resistedin oral cultures. When asked to explain what atree is, one oral peasant replied, “Why should I?Everyone knows what a tree is, they don’t needme telling them.” Ong’s response to this is,

Why deªne, when a real-life setting isinªnitely more satisfactory than a deªni-tion. . . . The peasant was right, there is noway to refute the world of primary orality. Allyou can do is walk away from it into literacy.(p. 53)

One might argue that the above questions werenot asked in the correct context. However, it

appears that there is no conceivably “correct”way to ask oral people such questions. In fact, inoral cultures, intelligence is something that ismeasured based on a person’s actions or skills,not by their response to verbal questions in atest: “Written examination questions came intogeneral use (in the West) only well after printhad worked its effects on consciousness, thou-sands of years after the invention of writing”(ibid., p. 55). Thus, asking abstract questions totest someone’s knowledge without any socialcontext is an alien concept in an oral culture.

It is important to note that just because oralpeople think situationally or concretely does notmean they are incapable of abstract thought.Indeed, there are complex layers of meaning inoral knowledge that are captured in idioms, folk-lore, mythology, riddles, and historical narratives.

Orality-grounded HCIDDrawing on Ong’s conceptualization of orality, wesuggest the framework of orality-grounded HCID. Bydrawing on a speciªc focus on oral end users, thisframework calls for a fundamental rethinking ofHCID both in terms of design principles and userresearch methodology. While the corollaries of theframework echo the sentiments of many user-centric methodologies popular in HCID research—such as user-centered design, participatory design,and participatory action research—there is still anew story to tell here in that the proposed frame-work attempts to explicitly take into considerationhow such users think, learn, and process informa-tion. Orality-grounded HCID is based on a testable,and falsiªable, model of oral users, and thus enablesHCID researchers to engage in more groundedresearch and design.

Below, we outline our proposal in terms of princi-ples of design and methodological considerations.These guidelines make explicit the fundamentaldifferences between mainstream HCI practice andorality-grounded HCID at both levels.

Principles for Interface and InformationDesignWhen designing interfaces for information access byoral users, several dimensions of orality are useful toconsider:

Information needs to be rooted in common ex-perience with speciªc examples. Abstractdescriptions, such as “breastfeeding should be con-

tinued even when the child has diarrhea” is not aseffective as describing a speciªc example of amother who is faced with this particular issue.Ideally, new information should be described interms of familiar cultural memes, and preferablyusing the culture’s own oral formulae. Thus, insteadof using a generic “mother,” the information shoulddraw on existing characters—perhaps a widelyknown maternal character in this case—in the com-munity’s folklore. Furthermore, drawing on a cul-ture’s oral formulae is best left to local members ofthe culture. IDE hired local storytellers to write andperform a play centered on a treadle pump (Polak,2008), which was an excellent strategy since localstorytellers understand their culture’s oral formulaebetter than anyone else.

Narrative stories are more memorable andmore effective at conveying information thanneutrally listed bullet points. Oral cultures donot have neutral lists. Ong wrote, “Even genealogiesout of such orally framed tradition are in effect com-monly narrative. Instead of a recitation of names,we ªnd a sequence of ‘begats,’ of statements ofwhat someone did: ‘Irad begat Mehajael, Mehjaelbegat Methusael, Methusael begat Lamech’ (Gene-sis 4:18)” (Ong, p. 98). Thus, listing bullet points ofinformation, such as how to improve cow yield, willnot work as well as telling the story of a farmer whoused a speciªc method and how he was able toincrease his yield (Gandhi et al., 2007). Moreover,dramatic descriptions will work better than neutralones. IDE created a full-length movie that showedthe value of a treadle pump through a dramaticstoryline (Polak, 2008), which was more effectivethan a neutral description of how such a pumpcould improve a farmer’s bottom line. The full-context video method (Medhi & Toyoma, 2007)leveraged the same principle.

Rhythm aids recall. IDE hired local troubadours tocompose a song about a treadle pump and to per-form it at farmers’ markets and fairs in Bangladesh(Polak, 2008). Even without the creation of an entiresong, content with rhyme and alliteration is likely tobe both understood and remembered by oral usersmore effectively than prose.

Linguistic style should be structured additively,not hierarchically. While it is common in literatematerial, using subordinative conjunctions such as“while,” “then,” “since,” and “although” is

uncommon in oral users’ communication. Compli-cated sentence structures impose a cognitive loadon the user. Instead, it is preferable to use coordi-nating conjunctions, such as “and,” “or,” and “so,”which do not create hierarchy.

Redundancy needs to be embedded in the con-tent. Redundancy is an important part of oral com-munication, mainly because of the ephemeral natureof speech. The user should also be given ampleopportunity to request repeated presentations ofcontent that has been given before. It should benoted, however, that explicitly requesting repetitionmay be less natural for the oral user than having thecorrect amount of redundancy already embeddedwithin the content. In the Digital Green project,researchers found that farmers would ask for thevideo content to be repeated multiple times; ourprediction is that a more optimal solution would beto design the video content to have redundancy“baked in” the content itself.

Each and every word needs to be understood.In an oral community that speaks one common lan-guage, there are no unfamiliar words, and oral usersnever face unknown words in daily life. Thus, evenone unfamiliar word can confuse the user com-pletely, and care should be taken to ensure that nosuch words exist in the system’s content. This iscommon when content comes from experts, whoare literate and have a more diverse linguistic back-ground than an oral person. This strongly suggeststhat locally generated content—as in the case ofthe Digital Green project—is likely to be the mosteffective option in many cases, in terms of its rele-vance and understandability.

Abstract categories should be avoided. Oralusers do not categorize the same way as literateusers, and do not think in terms of abstract catego-ries (Luria, 1976). The use of categories should beminimized and if their use is essential, it should bekept in mind that the designer’s choice of categorieswill most likely not match the expectations of theuser. Also, hierarchies in information architectureshould be avoided. Browsing multiple depths ofinformation (e.g., as in a Web page, or when usingnavigation metaphors of “up a level”) is difªcult fororal users (Deo et al., 2004).

Requiring adherence to speciªc spoken wordsor phrases is less likely to succeed. Oral usersperceive speech as a continuous stream, rather than





as discrete words. In our work with speech recogni-tion interfaces, one of the most common complaintsof oral users is that they are afraid that they will“say the wrong thing”; they understand they haveto speak in individual words, an action that is unnat-ural for them, and their comments reºect their dis-comfort with this requirement. While it is technicallydifªcult, it may be necessary to widen the speechrecognizer’s grammar to allow for more naturalutterances, rather than forcing the user to speak insingle word commands.

Oral people do not internalize new informationthe same way as literate people. Since internal-izing new information is comparatively expensive fororal people—they cannot ofºoad their memoryrequirements onto the technology of writing—itappears that they are more selective when choosingwhether to internalize new information. In ourresearch, we have seen many times that even whenusers understand each word in the content perfectly,when asked a question on that content, users willrespond based on prior knowledge. In a study weconducted with health workers of varying literacylevels, even after reading a speciªc paragraph intheir manual ºuently, their response to a questionon that content was the opposite of what it said,based on prior knowledge (Sherwani et al., 2007).Similar effects have been found in other research onoral people (Luria, 1976) and poor readers (Doak,1995), and have also been anecdotally reported byother HCID researchers (Etienne Barnard, personalcommunication). This might happen because theformal content is not seen as relevant, or it isn’torganized in the same linguistic style that usersexpect (e.g. narrative, concrete examples rather thanneutral and abstract information), or for other rea-sons; however, it is important in the design of aninterface to know that this is a problem that needsto be engaged.

Oral people give more importance to thesource of information than literate people.Writing establishes “context-free language” (Ong,1982), where information is not linked to any partic-ular source. For oral people, however, all informa-tion is social and traceable to a person. Whenfarmers were shown videos of other farmers dem-onstrating agricultural best practices, the most com-mon question the viewers asked was what thedemonstrating farmer’s name was, and which villagehe was from (Gandhi et al., 2007). This may also

explain why source-neutral information—of theform presented in a book or in a persona-less sys-tem—is not internalized: It is missing the essentialfeature of social context, not only of the informa-tion’s relevance to the real world, but also its trace-ability to a trustworthy human being. Thus, systems,such as Digital Green, that provide informationthrough people perceived to be trustworthy arelikely to be more successful than systems that trans-mit information without this feature.

Given all the recommendations listed above, itbecomes apparent that information designed for lit-erate users—virtually all written material—is notappropriate for oral consumption. Oral and literateusers require content with different organization,presentation, and context.

Furthermore, end users may be the best resourcefor content creation and content adaptation. Liter-ate people produce both written and spoken con-tent optimized for literate people (Ong, 1982,p. 56). Even “human access points” (Marsden et al.,2008), or literate technologists from the local com-munity, may not understand how to effectively altercontent for optimal consumption by oral users,although they would do a better job than a naïve,non-local designer. However, it may be the case thatinvolving users is essential for content creation andcontent adaptation. The success of the Digital Greenproject is a case in point; it doesn’t just involve usersin the creation of the content, but features them asthe star performers in the “Farmer Idol” videos thatform the backbone of the system.

Methodologies for User ResearchEven more basic than principles for design, method-ologies for conducting user research must also beadapted for application within oral cultures. We canlearn from the signiªcant number of successful eval-uations of designed artifacts in the context of oralcultures, projects such as Digital Green (Gandhi etal., 2007), eSagu (Reddy et al., 2007) and WaranaUnwired (Veeraraghavan et al., 2007). In these andother related efforts, the common experience hasbeen that standard user studies have various layersof problems. Ultimately, it may be the case that afundamentally different method of evaluation needsto be explored for the HCID context. In this section,we explore the issues with existing methodologiesand make some suggestions for how the methodol-ogies can be adapted in light of the differencesbetween oral and literate cultures.



Overall, standard user study practice needs tobe considerably rethought for application inoral contexts. Even though literate participantsmay ªnd user studies a novel concept, they are likelyto have experienced the individual components of auser study in school, and would likely consider themsomewhat familiar. For oral people, a user study isan alien experience. The various steps involved in auser study—the facilitator reading out instructionsto the participant, teaching the steps in the task,and then asking the participant to perform it them-selves to answer examination-style questions—is aclinical abstraction that is alien to the lifeworld of atypical oral person. Oral people do not think andlearn in the way that user studies expect them to,and they are not used to being asked abstract, con-text-free questions. Thus, it is arguable whetherresults from such studies are of much analyticalvalue in oral contexts. Further work is needed toexplore alternative methods for user studies adaptedto the HCID context.

The experimenter always has a higher socio-economic status than the participant and isusually foreign to the local community, leadingto feelings of intimidation and “performanceanxiety.” This is often an overlooked feature ofHCID user studies and deserves a fundamentalrethinking of user study roles. Recruiting and train-ing local user study facilitators from the communityhave worked well in our experience. Moreover,involving the facilitator in the motivation and designof the user study is useful for a number of reasons,as it increases their ownership of the process, andprovides them the necessary context for why eachstep of the user study is important, in addition tolisting the actual steps they need to perform.

When presenting a novel interface, the contextin which it is to be used and the motivation forthe system should be presented as concretelyand vividly as possible. Given oral users’ prefer-ence for situationally grounded examples relevant totheir lifeworld, it is imperative to present the systemas a solution to a problem that users deem impor-tant and for which users perceive current solutionsas inadequate. Whenever possible, the systemshould be introduced as a solution to a widelyremembered, speciªc instance of a local problem.For example, in health care, there may be a recentstory of a person affected by an illness that no oneknew how to cure, which everyone remembers viv-

idly, perhaps because of its tragic nature. By askinglocals about such examples beforehand and thenportraying a health information access system withina narrative that ties into the local example of aproblem, as the potential solution, users may seethe social context of the system more clearly thanthey would if it were to be given as a solution to theproblem in general (e.g., “Imagine if a person inyour village became ill . . .”). That approach is stillbetter than a generic solution in the abstract (with-out a narrative, e.g., “The system can be used togive health information when the need arises”). Amotivation presented with a narrative, tied to thespeciªc local folklore of the community, could makethe intervention seem more relevant to the needs ofthe community.

This suggests why the full-context video method(Medhi & Toyama, 2007) works so well: It creates adramatized and visual example of how the interfacecould be useful, instead of introducing it in abstract.Given the need to localize all content, it is unclearwhether a video would be more cost-effective atproviding context than other alternatives (e.g., face-to-face interaction) with a narrative that is custom-ized to the speciªc experiences and memory of theuser or community. Given the high success rate ofthe full-context video method (100% of the usersexposed to the video were able to successfully com-plete the task, compared to 6% in the controlgroup), full-context video is clearly a winningmethod, and may be the optimal choice.

Oral participants do not remember neutral userstudy tasks, and do not actively engage withthem. It is a known issue that artiªcial scenariosand ªctitious needs are unlikely to be internalized byparticipants. For oral users, tasks cannot be pre-sented textually—and so cannot be reviewed atwill—and furthermore, if the task is abstract andneutral, which is standard HCI methodology, it isunlikely that the task will be remembered, much lessinternalized. Engaging with a task that one does notremember or believe in is also unlikely. The Bolly-wood method (Chavan in Shaffer, 2004) is an excel-lent solution to this problem. In this method, userstudy tasks are couched in a dramatized narrative,which leads to more involved and interactive partici-pants, who would otherwise not contribute anyfeedback when faced with neutrally presented tasks.Thus, oral users are more likely to remember andengage with a dramatic narrative than with an



abstract and neutral request for information. It isunclear how much of the difference is attributableto (a) participants’ improved memory of the drama-tized tasks, (b) their appreciation for these tasks’wider social context and relevance to the humanlifeworld (an issue of motivation), and (c) their per-ceived freedom to critique when in role play (whatChavan attributes its success to), or other factors.Further work is needed to explore what aspects ofthis method work well, and why.

Giving user study participants descriptive in-structions on how to use the system is notsufªcient. Oral people do not learn by verbalizedinstruction—rather, they learn through apprentice-ship and practice. This is neatly analogous to the dif-ference between declarative and proceduralknowledge in the ªeld of artiªcial intelligence. Inour own work, we have found that a mentor-apprentice model, in which a facilitator does aguided walkthrough of a sample task with a partici-pant, works well. Also, participants’ understandingof the system improves further with an“incrementally-removed training wheels”-approachwith multiple practice tasks, where explicit help isprovided on each step for the ªrst sample task, andis gradually reduced for each subsequent practicetask, until no help is provided on the ªnal sampletask unless required. The approach used in the full-context video method is different, and we hypothe-size that oral users would learn better with a per-sonal, mentored learning experience than bywatching a video demonstrating another person’suse of a system. Further work is needed to ascertainthe best strategy for teaching an interface to theuser—both for the design of the system and forevaluating the system in a user study.

Likert Scales are an inappropriate measurementtool for oral people. A Likert Scale question con-sists of an abstract statement, such as “I found thissystem easy to use,” and a list of numbered options,usually 1 through 7, with 1 representing “stronglydisagree” and the highest number (e.g., 7) repre-senting “strongly agree.” Participants are asked toselect the option that most closely describes theirlevel of agreement with the given statement. Whenused with oral people, this tool is problematic for anumber of reasons:

• The Likert Scales method requires that the op-tions to be written down and presented textu-

ally; however, oral people cannot read and ver-bal presentation of Likert Scales has not beenrigorously proven as a valid methodology, eventhough they might be used in practice.

• The concept of “context-free” abstract state-ments is a literate construct. Oral people donot conceive of statements in this form—therehas to be someone who said any given state-ment. Moreover, who this someone is makes adifference as to how likely they are to agreewith the statement; issues of credibility, trust,authority, politeness, political considerations,etc., are involved.

• Categorizing agreement into discrete chunks isalso a literate construct. Oral people do notthink in terms of such categories, and we haveseen countless examples of oral user study par-ticipants confused at such categorization.

• Praise and criticism are often given indirectly inoral cultures, and thus asking participants todirectly praise or critique an artifact with theirsubjective evaluation might not align with par-ticipants’ expectations.

In our own work, we have had signiªcant prob-lems with Likert Scales, where participants oftengave polite, positive answers, which on deeperexamination and triangulation were not representa-tive of their true opinions. In one user study we con-ducted in rural Pakistan, oral participants wereunable to speak the language of the speech inter-face they were using and could not correctly answerany of the health information questions they wereasked to learn from the system. Yet, when asked torate the system in terms of various dimensions—including ease-of-use and understandability—theyrated it very highly (Sherwani et al., 2009). Similarexperiences have been reported by other researchersin South Africa (Aditi Grover, personalcommunication).

The “auto-rick radio” method (Chavan, 2007) pro-vides an interesting alternative to Likert Scales. Insteadof representing a subjective evaluation on a linearscale, participants are asked to twist a knob, shapedlike a radio’s volume knob, to represent their feelingson a particular dimension. Chavan wrote that:

For these users, the concept of a difference in de-gree (moving from negative to positive) being rep-resented by a horizontal straight line seemed veryconºicting. The feeling was that if the different



points in the scale represented different degreesof an attribute, then they could not appear to beon the same level, as they did with the straighthorizontal line. . . . Hence a knob control was de-vised which resembled the volume control knobof the radio that all users were very familiar with.(Chavan, p. 13)

However, this explanation confounds various issues:the difªculty in quantifying emotion for oral users, thedifªculty in relating emotion to spatial terms, and thedifªculty of expressing their choice using a novel inter-face. By providing a familiar interface (the radio vol-ume knob), participants may have found it easier toexpress themselves, but whether their expression cor-rectly represented their subjective emotions remains tobe seen. Rigorous methodological research is requiredto explore these issues in detail.

One potential alternative to Likert Scales in gen-eral is to instead solicit open-ended answers andhave the experimenter decide on the appropriatecategory—after all, what is the point of asking aparticipant to do something, like quantify an emo-tion, that that person really cannot do? Chavan sug-gested “using storytelling to ªnd dissonances” as ameans of asking users about their experience with aparticular product (ibid.); however, this methodcould be extended to subjective measurement inwhich participants could be asked, for example, tonarrate instances where they had trouble perform-ing the user study task, as an alternative to askingthem to subjectively rate its difªculty. The experi-menter could use the quantity and quality of theseinstances to determine a quantiªcation of the user’sexperience, if desired. However, if the participantand experimenter are both present in the sameroom during the study, the participant would likelywonder why the experimenter is asking such ques-tions, since surely the experimenter saw the partici-pant’s performance and should obviously then knowthe answers. As such, care would need to be takento set up the scenario appropriately—perhaps byhaving a different experimenter conduct the subjec-tive evaluation—to ensure that the participant viewsthese questions as valid and worthy of a realresponse.

Thus, it can be seen that standard user studieshave various layers of problems, and while theabove discussion points out some suggestions forhow the methodologies can be adapted, it may bethe case that a fundamentally different method ofevaluation needs to be explored for the HCID con-text. Alternatively, testing deployed interfaces “in

the wild” should yield more representative resultsthan user studies would, and more so than in liter-ate contexts. There are a signiªcant number ofresearch projects designed this way, such as DigitalGreen (Gandhi et al., 2007), eSagu (Reddy et al.,2007) and Warana Unwired (Veeraragavan et al.,2007), and this may be a better, though in mostcases more difªcult, method of evaluation.

Conclusion and Future WorkIn this paper, we have argued that because a dispro-portionate amount of HCI practice to date hasfocused on literate users, it is important to under-stand that standard HCI practice is not always appli-cable in oral contexts. Orality theory provides aunique lens with which to understand oral users,and we have attempted to synthesize a frameworkthat provides guidelines and testable predictions fordesign and evaluation with such contexts. We havealso discussed best practices in HCID based on exist-ing research in the context of the framework. Muchwork is needed, however, to substantiate, qualify,verify, refute, and build on this work. ■

ReferencesAckermann, E. K., & Decortis, F. (2007). Kids story

“writers”: POGO, tell-tale, sprite. Proceedings ofthe 6th International Conference on InteractionDesign and Children, pp. 161–162, Aalborg,Denmark.

Bergstrom, T., & Karahalios, K. (2009). Conversationclusters: Grouping conversation topics throughhuman-computer dialogue. Proceedings of the27th International Conference on Human Factorsin Computing Systems, pp. 2349–2352, Boston,MA, USA.

Chavan, A. (2007). Around the world with 14 meth-ods. Human Factors International. White paperretrieved on October 10, 2008, from http://www.humanfactors.com/downloads/whitepapers.asp#bollywood

Conrad, E. (2003). Aether: Reading with tactile vi-sion. ACM SIGGRAPH 2003 Sketches & Applica-tions, p. 1, San Diego, CA, USA.

Davis, M. (1997). Garage cinema and the future ofmedia technology. Communications of the ACM,40(2), pp. 42–48, New York, NY, USA.

Deo, S., Nichols, D., Cunningham, S., & Witten, I.

(2004). Digital library access for illiterate users.Proceedings of 2004 International Research Con-ference on Innovations in Information Technol-ogy., pp. 506–516, Dubai, UAE.

Doak, C., Doak, L., & Root, J. (1996). Teaching pa-tients with low literacy skills. Retrieved on Sep-tember 10, 2008, from http://www.hsph.harvard.edu/healthliteracy/doak.html

Fischer, G. (2006). Distributed intelligence: Extendingthe power of the unaided, individual humanmind. Proceedings of The Working Conferenceon Advanced Visual Interfaces, pp. 7–14,Venezia, Italy.

Huenerfauth, M. (2002). Developing design recom-mendations for computer interfaces accessible toilliterate users. Master’s thesis, University CollegeDublin, Ireland.

Gandhi, R., Veeraraghavan, R., Toyama, K., &Ramprasad, V. (2007). Digital green: Participatoryvideo for agricultural extension. Conference pro-ceedings of ICTD 2007, pp. 21–30, Bangalore,India.

Grudin, J. (2007). Living without parental controls:The future of HCI. Interactions, 14(2), 48–52,New York, NY, USA.

Luria, A. (1976). Cognitive development: Its culturaland social foundations. Cambridge, MA: HarvardUniversity Press.

Marsden, G., Maunder, A., & Parker, M. (2008). Peo-ple are people, but technology is not technology.Philosophical transactions of the Royal Society,A:1–10, London, UK.

Mathur, P., & Karahalios, K. (2009). Visualizing re-mote voice conversations. Proceedings from the27th International Conference on Human Factorsin Computing Systems, pp. 4675–4680, Boston,MA, USA.

Medhi, I., Sagar, A., & Toyama, K. (2006). Text-freeuser interfaces for illiterate and semi-literate us-ers. Proceedings from ICTD 2006: InternationalConference on Information and CommunicationTechnologies and Development, pp. 72–82,Berkeley, CA, USA.

Medhi, I., & Toyama, K. (2007). Full-context videosfor ªrst-time, non-literate PC users. Paper pre-sented at IEEE/ACM International Conference on

Information and Communication Technologiesand Development. Bangalore, India.

Ong, W. (1982). Orality and literacy: The technol-ogizing of the word. London: Routledge.

Parikh, T., Ghosh, K., & Chavan, A. (2003). Designstudies for a ªnancial management system formicro-credit groups in rural India. Proceedings ofthe 2003 Conference on Universal Usability,pp. 15–22, Vancouver, BC, Canada.

Polak, P. (2008). Out of poverty. Berrett-Koehler Pub-lishers. San Francisco, CA, USA.

Reddy, P., Ramaraju, G., & Reddy, G. (2007).eSagu™: A data warehouse–enabled personal-ized agricultural advisory system. Paper presentedat 2007 ACM SIGMOD International Conferenceon Management of Data, pp. 910–914, Beijing,China.

Shaffer, E. (2004). Institutionalization of usability.Addison Wesley Professional. Boston, MA, USA.

Sherwani, J., Ali, N., Mirza, S., Fatma, A., Memon,Y., Karim, M., et al. (2007). HealthLine: Speech-based access to health information by low-literate users. Proceedings for 2007 IEEE/ACM In-ternational Conference on Information and Com-munication Technologies and Development.,pp. 131–139, Bangalore, India.

Sherwani, J., Palijo, S., Mirza, S., Ahmed, T., Ali, N.,& Rosenfeld, R. (2009). Speech vs. touch-tone:Telephony interfaces for information access bylow literate users. Proceedings for ICTD 2009: In-formation and Communications Technologies andDevelopment, pp. 447–457, Doha, Qatar.

Stahl, G. (1999). Reºections on WebGuide: Seven is-sues for the next generation of collaborativeknowledge-building environments. Proceedingsof the 1999 Conference on Computer Supportfor Collaborative Learning, Article 73, Palo Alto,CA, USA.

Veeraraghavan, R., Yasodhar, N., & Toyama, K.(2007). Warana unwired: Replacing PCs with mo-bile phones in a rural sugarcane cooperative. Pa-per presented at ICTD 2007: Information andCommunication Technologies and Development,pp. 89–98, Bangalore, India.

Wright, A. (2008). Primal Actions, Interactions,15(1), 11–12, New York, NY, USA.



Documents

orality-hcid-itid09