Ivy League IVR
Respect-it can be a difficult, yet highly desirable value to attain. Most people want respect; it's a basic human motivation to be treated with dignity. It's also easy to know when we have been treated with disrespect-we may feel tense, frustrated, angry. Respect is something that you can observe and infer. And yet, it is seldom discussed as an important component of a speech system. In fact, some might find it laughable to think that a machine would ever show respect or be perceived as respectful. Is it important? Should designers be concerned about an abstract, decidedly human characteristic like respect? If designers are interested in providing high-quality customer service, research suggests that they should.
Respect and Customer Service
Speech technology is often used as a form of e-service or technology-based customer service. As such, it has a relationship to other forms of customer service, such as that provided on the Web, or through service providers like salespeople, service representatives, store clerks, and restaurant wait staff. But the person-or personality-in a service role usually has a significant impact on the customer's experience with the business. Research on human-to-human customer service suggests that the personality of the service provider plays an important role in determining the quality of service provided.
In their broad review of studies measuring service provider personality, researchers found that "conscientiousness" (competence and willingness to perform) was most strongly related to service quality, but so was "agreeableness" (trust, modesty, compliance), which would support a respectful attitude toward a customer1. Other research has shown that excellent interpersonal skills are also positively related to service performance and, even more importantly, customer attitude about the service received. Service providers who are outgoing and have the ability to negotiate, get along with others easily, and execute their roles based on customer needs are most successful in service jobs. A study from 2001 showed that assertiveness or deference to a customer determined whether a customer was satisfied with a service. If the service provider was aggressive, the customer showed decreased satisfaction with the service. However, if the provider was very polite and respectful, the customer was satisfied2.
The next time you go to a store, restaurant, or any other establishment that provides services, consider that respectful treatment is a basic desire we all have. Some researchers suggest that service interactions are a special form of general interpersonal interaction between people. For a service encounter to go smoothly, both the customer and the service provider must adopt coordinating, yet complementary, roles that are clearly defined in our collective understanding3. Thus, demonstration of respect toward a customer is one common expectation of any customer service and underlies judgments of service quality, provokes an emotional response in the customer, and ultimately determines the customer's intention to become a loyal consumer of a business. In fact, researchers have argued that customer satisfaction is the result of service providers and customers sticking closely to their expected roles; it's only when they veer away from them (by displaying unexpected behavior) that dissatisfaction results.
So, if we want to show respect for a customer, what are the behaviors involved? Interestingly, a major way respect is demonstrated is through the use of speech and language, which makes it relevant for speech interface design. Speech act and politeness theory4 suggest that the choice of phrase indicates a level of respect for our conversational partners. According to these theories, communication has three components: the meaning of an utterance, the intent of the speaker, and the effect on the hearer. Therefore, much of what we communicate lies in how we say things, rather than specifically what we say. Consider the following examples:
1. Say your credit card number now.
2. What's your credit card number?
3. Could you please tell me your credit card number?
With each utterance, the speaker is requesting specific information, specifically a credit card number. However, the particular form of language conveys much more. In the first statement, the speaker is asserting his own desire for and ability to control the information exchange, while the two questions are less assertive and the third example even questions whether the credit card holder is willing to provide the information. Unlike the first command, the other two utterances suggest more choice for hearers to actually provide the information, which indicates that they have more control of the information than the speaker. In this sense, the speaker indirectly conveys more respect for the conversational partners' status through the choice of syntax.
Communication theory also suggests that many of our utterances are designed to balance our own self-respect with demonstrating respect toward others. We want to show that our partners are important, but we also want to make ourselves seem as if we have importance too. Think about individuals who you speak with during the day: do some of them come across as particularly assertive or domineering when they speak? Do others seem particularly passive or deferential? What about people who have a different status than you do, like a manager or parent? In addition to choosing different linguistic forms like statements or questions, individuals who perceive that they have greater power are more likely to initiate conversation or interrupt, speak over others, and speak longer (hold the floor). People who have lesser power are more likely to add tag questions (e.g., "I thought that was fun yesterday, didn't you?") use passive voice, add modifying verbs (e.g., "I might like to do it" instead of "I like to do it"), or have a more difficult time obtaining a speaking turn. An interesting self-experiment is to observe the patterns of a variety of individuals and notice how they use communication skills in conversation. In addition, consider your own conversational approach with these partners. How does your language change when you want to demonstrate respect for your partner, especially if that person has a higher rank or status than you? What language choices make you feel like your conversational partners are being respectful to you?
Although these social interaction processes are interesting in their own right, they become especially so when one of the conversational partners is a machine. In this case, does the machine acquire status or power in relation to a human partner? Logically, perhaps the answer is no, but when the machine uses conversational markers that imply power, the psychological impact on the user/partner is similar to that in a more typical conversation. The perception of power occurs because our responses to speech and language are automatic and beyond our rational understanding of the interaction. In addition, we've been reacting to these markers ever since we learned the subtle aspects of language use in childhood, so the language behaviors we observe have had their social meaning for a very long time. Let's compare two brief, partial interactions (See Figure 1).
Notice that in Interaction 1, the system uses at least two markers indicating its greater power and status over the human user: it uses statements (commands) to elicit information and holds the floor on a topic of its own selection (a new offer from the bank). In the second interaction, these markers are substituted with different linguistic characteristics that imply more status for the user; specifically, the system uses questions and shorter, more equivalent turns. Imagine if each of these interactions continued in their current patterns. Which one subtly conveys to the user a more respectful attitude, in line with the users' expectation of how they should be treated? Which one is more likely to be seen as adhering to the appropriate customer service role?
It's clear that interface designers make choices about the linguistic behaviors of a speech e-service system. But just because particular behaviors are present doesn't mean the human user will show preferences or awareness of them. Or does it? Do users actually know whether or not they've been treated with politeness and respect? If you think back to your conversations today, you'll quickly realize that you know when you have been treated appropriately and when you haven't, but you might have a harder time pointing out exactly what makes you feel this way. Similarly, data indicates that users may very well be aware of a decidedly human-like trait like respectfulness in a speech system. Even more important, this perception may also be associated with other outcomes of speech systems, including repeat usage, call retention in the system, and ultimately, operational savings.
Little research has been done on this topic for speech user interfaces. Respect is a complex variable that requires multiple items on rating scales to adequately assess it. However, a recent study showed that statements associated with respectfulness, such as, "The system made me feel like I was in control," "The system used terms I am familiar with," "The system seemed polite," "The system seemed courteous," and "The system spoke at a pace that was easy to follow" are important to ratings of usability in speech interfaces5. Interfaces that linguistics and psychology experts rated as high quality also received high ratings (closer to 7) on these five statements (see Figure 2). Other findings suggest that ratings of respectful behavior are associated with emotional outcomes like satisfaction. Using correlation analysis, high ratings of control were most strongly associated with high satisfaction (correlations - r=0.53, significance levels - p<0.001), followed by perceptions of an organized interface (r=0.50, p<0.001). The other ratings of politeness, courteous behavior, and pacing were also positively correlated with satisfaction, but not as strongly (r=0.30 to 0.40, p<0.0001).
In contrast, participants who gave high ratings on negative items like "The system gave me more details than I needed" and "I felt like I had to wait too long for the system to stop talking so I could say something" for low quality interfaces (judged by an expert panel) also rated greater dissatisfaction with a speech interface (r=-0.10 and -0.36, p<0.003, respectively). Thus, this research suggests that, just as in human-to-human customer service, speech and language choices in a speech interface can also produce measurable differences in users' perceptions of abstract qualities like respect.
Designing with Respect: Why Not?
Research suggests that making design choices to show respect for the customer will create better outcomes, but in the practical world, actually making these choices can be much more problematic. Because respectfulness does implicate several aspects of an interface design (including its overall organization, linguistic style of prompts, approach to error recovery, and presentation style of a voice talent), designers may need to carefully juggle these considerations against a wide variety of competing requirements. There are a number of issues that can derail even the best of intentions to create highly usable and respectful systems. These issues include the following:
1. Minimal respect
Several service marketing researchers lament that many businesses still minimally seek meaningful customer input on their services. To this end, some business cultures seem to buy into a "if we offer it, they will come" mentality. In situations where customer service is not seen as a differentiator, the desire to create an outstanding user experience can be either nonexistent or secondary to other outcomes such as call retention or operational savings. If other modes of customer service are difficult to use, inconsistent, or unpleasant, it's very likely a speech interface will follow the problematic service approach. Ironically, that a pleasant, usable interface is a necessary prerequisite for other financial outcomes can often fall on deaf ears.
2. Complex processes or logic
Many companies have internal processes that have evolved over time and flow through a variety of teams and structures within the organization. Often, individuals may have knowledge of only some parts of these complex processes. Creating a simple, understandable speech interface design can be especially problematic in these situations, because the business logic underlying the dialogue suddenly becomes transparent to the user. Because the user has no knowledge of the history and reasoning that created this logic, it is unlikely that user's own reasoning about what is involved in executing a task with the system will match the actual logic of the system.
3. Technical limitations
In addition to complex business logic, many companies also have technology infrastructures that have evolved over time. Pieces of data may reside in very different locations and may have varying levels of accessibility. Thus, the technology itself may force a particular sequence of data input or output, which impacts the sequence of turns in a conversational interaction. In a case where this sequence is similar to the user's expected sequence of information exchange, this issue will not pose a problem; however, when the technical requirements and user expectations for data sequencing differ, usability problems will result.
4. Narrow customer focus
As noted earlier, many companies have significantly more knowledge of their own internal organization, business processes, and technical infrastructure than they do about their customer. In practical experience, the number of companies that are able to provide highly detailed demographic data on their customers (user population) and their needs is relatively few. Even fewer track and mine their customer service data to have an ongoing view of how customers want to interact with the business. If there is a significant mismatch in business and customer knowledge, stakeholders can make the mistake of requiring an interface design that makes sense to them, but not the user.
5. Attitude barriers
Even if companies recognize that an interaction will be difficult or problematic for their users, they may not be willing to make significant internal changes for the sake of usability. Certainly, many of the changes required can be time consuming, expensive, or difficult to execute. Many companies go into a speech application project assuming that it's a simple thing to build and don't realize that major impediments can get in the way of that "simple" dialogue. A related problem is legal issues in customer communication - many companies have lengthy disclaimers and legal notices that they provide to their customers. Often, these messages use the complex sentence structure and advanced terminology that might be appropriate for a corporate legal team, but not John Q. Public. The net effect of these messages is often confusion or even a perception that a company is more concerned about protecting itself than providing high-quality services. A commitment to speech technology success may often mean that a company must commit to business process and technology innovation as well.
6. Not focused on interface or outcome
Finally, in other cases, stakeholders can have strict requirements about their desired business outcomes, such as a specific incremental increase in call retention. However, they may not fully appreciate the strong relationship between this goal and the quality of the user interface. Excellent user-centered design incorporates process, as well as its outcome. In addition, having skilled resources on the team who understand human conversational behavior, cognition, and social interaction is vital to the success of an interface. Communication is humans' most complex skill and designing a successful communication experience is the result of knowledge, experience, science, art, and advocacy.
These issues, and others like them, can influence the development process, and in turn affect user interface design, the customer service experience, and desired business goals for an application. Realistically, in every speech application project, the relevance of each issue will vary and others may also exert themselves.
In many cases, the best way to alleviate these risks is to be sure that projects are supported by strong, experienced, and knowledgeable designers who can see user needs and impacts with an eye unblemished by organizational and technical concerns. In addition, all members of the development team need to be aware of their own biases and limitations, and allow time to come to a data-based understanding of the issues that could negatively impact the success of an interface. Certainly, allowing adequate time and resources is not easy to accomplish in our fast-paced business environment. But, as in a good speech interface, respectfulness can be one key to negotiating complex interactions and achieving intended success. In the end, understanding and designing for respectfulness can pay great dividends in automated customer service.
Melanie D. Polkosky is a social-cognitive psychologist and speech-language pathologist who has researched and designed assistive and enterprise communication technologies for over 10 years. She is currently a senior human factors psychologist for IBM Conversational Solutions Group, with expertise in social cognition, interpersonal communication, usability measurement, and social impacts of technology.
1. Mount, M., Barrick, M., & Stewart, G. (1998). Five-factor model of personality and performance in jobs involving interpersonal interactions. Human Performance, 11(2/3), 145-165.
2. Yagil, D. (2001). Ingratiation and assertiveness in the service provider-customer dyad. Journal of Service Research, 3(4), 345-353.
3. Solomon, M., Surprenant, C., Czepeil, J., & Gutman, G. (1985). A role theory perspective on dyadic interactions: The service encounter. Journal of Marketing, 49, 99-111.
4. Austin, J. (1962). How to do things with words. Oxford: Clarendon Press. Brown, P., & Levinson, S. (1987). Politeness: Some universals in language usage. Cambridge: Cambridge UP. Searle, J. (1969). Speech acts. Cambridge: Cambridge UP.
5. Polkosky, M. (2005, in press). Toward a social-cognitive psychology of speech technology: Affective responses to speechbased e-service (Doctoral dissertation, University of South Florida, 2005). Dissertation Abstracts International.