Cookie Preference Centre

Your Privacy
Strictly Necessary Cookies
Performance Cookies
Functional Cookies
Targeting Cookies

Your Privacy

When you visit any web site, it may store or retrieve information on your browser, mostly in the form of cookies. This information might be about you, your preferences, your device or used to make the site work as you expect it to. The information does not usually identify you directly, but it can give you a more personalized web experience. You can choose not to allow some types of cookies. Click on the different category headings to find out more and change our default settings. However, you should know that blocking some types of cookies may impact your experience on the site and the services we are able to offer.

Strictly Necessary Cookies

These cookies are necessary for the website to function and cannot be switched off in our systems. They are usually only set in response to actions made by you which amount to a request for services, such as setting your privacy preferences, logging in or filling in forms. You can set your browser to block or alert you about these cookies, but some parts of the site may not work then.

Cookies used

ContactCenterWorld.com

Performance Cookies

These cookies allow us to count visits and traffic sources, so we can measure and improve the performance of our site. They help us know which pages are the most and least popular and see how visitors move around the site. All information these cookies collect is aggregated and therefore anonymous. If you do not allow these cookies, we will not know when you have visited our site.

Cookies used

Google Analytics

Functional Cookies

These cookies allow the provision of enhance functionality and personalization, such as videos and live chats. They may be set by us or by third party providers whose services we have added to our pages. If you do not allow these cookies, then some or all of these functionalities may not function properly.

Cookies used

Twitter

Facebook

LinkedIn

Targeting Cookies

These cookies are set through our site by our advertising partners. They may be used by those companies to build a profile of your interests and show you relevant ads on other sites. They work by uniquely identifying your browser and device. If you do not allow these cookies, you will not experience our targeted advertising across different websites.

Cookies used

LinkedIn

This site uses cookies and other tracking technologies to assist with navigation and your ability to provide feedback, analyse your use of our products and services, assist with our promotional and marketing efforts, and provide content from third parties

OK
[HIDE]

Here are some suggested Connections for you! - Log in to start networking.

EXECUTIVE MEMBER
Ikhwal Sidiq
Assistant Manager Trade and Remittance Services
408
MEMBER
Richard Roberts
Adviser and Consultant
18
EXECUTIVE MEMBER
Selin İcer
Quality - Training & Academy Director
29
MEMBER
Thamer Noori
Director of Industrial Security and Safety Dept.
13

Article : Voice User Interface Visions

At first sight, a yes/no question seems like a simple dialog state for a voice user interface designer to craft. The possible responses are just yes or no, right? Well, no, actually. As with many issues in dialog design, surface simplicity masks a deeper complexity. This article describes the complexity of confirmation states, and outlines strategies recommended to improve the caller's experience.

Looking at data from many of our deployed applications, we observe that while a large number of users do respond to confirmation questions with yes, no, or one of their colloquial synonyms (such as yup, that's right or no it isn't), a significant number don't. For example:

System: Is there anything else I can help you with?
User: Yes. I'd like to make a payment.

As within the context of many live conversations, the user goes beyond yes to specify what they'd specifically like help with. Frequently, users even omit the yes (only specifying the next task). In one of our customer service applications, more than 20% of responses to this question fell into this category.

To provide the best user interface, the voice user interface designer needs to handle this type of behavior using a carefully deliberate combination of grammar coverage, effective prompting, and error handling, plus additional steps to avoid endless loops and improve the time to task completion.

Grammar Coverage
The first step in handling complex answers to yes/no questions is to expand the grammar coverage. In the example above, the grammar should also allow contextually logical phrases such as 'I'd like to make a payment', plus combinations like yes and no followed by such phrases. The exact combinations will depend on the context. In the case above, the grammar should not include phrases like "No. I'd like to make a payment", to avoid over-generation of grammars. In another case, though, different combinations should be allowed.


Dr. Lisa Guerra
VP Of Engineering
BeVocal


Dr. Ryan Bush
Senior User Interface
Design Lead
BeVocal


System: Fremont, California. Is that right?
User: No. Fulton, California.

Here, it makes sense to say no and then give an alternate value, but it wouldn't make sense to say something like "Yes. Fulton, California." Once again, the exact design of the grammar is very case-specific and depends on the context.

Effective Prompting
Once the grammar is broad enough to accept the users' likely utterances, we need to consider the prompting. If the system recognizes more than yes and no, it is tempting to let the user know about all the possible things they can say. For example, the confirmation prompt might say "Fremont, California. Say 'yes', or name another city and state." Or, "Fremont, California. Or, name another city and state." After all, we should let everyone know about the handy shortcut of naming another city and state, shouldn't we?

In our experience, no. The problem is that those prompts place too large a cognitive burden on the user. The prompts ask the user to choose from different types of actions: either saying yes, or naming another city and state. In normal conversation, we are ordinarily given choices that are of the same type but representing logical opposites: yes or no, for here or to go, apples or oranges.

In one application that used this type of prompt, we found an increased number of no speech errors (where the user just didn't say anything), fewer users saying 'yes', and other evidence of user confusion. In fact, even when users were presented with the correct city and state, many users repeated that city, instead of just saying 'yes':

System: Fremont, California. Say 'yes', or name another city and state.
User: Fremont, California.
System: Fremont, California. Say 'yes', or name another city and state.

This is definitely not the desired behavior. It increases time to task completion and reduces user satisfaction. So, sometimes simpler is better. We changed the prompt to "Fremont, California. Is that right?" and user confusion was measurably reduced. There was a slight decrease in the number of users that named another city and state (with a corresponding increase in no responses), but it is natural enough to name another city and state in that context that many users do it without prompting.

Error Handling
With the grammar and prompts in order, the remaining piece is error handling. If we're not careful, the user can get stuck in an unpleasant loop:

System: Getting movies for Fulton, California. Is that right?
User: Fremont.
System: Sorry, I didn't understand. Fulton, California. Is that right?
User: Fremont.
System: I still didn't understand. If you want Fulton, California, say 'yes'. If not, say 'no'.

In this example, the user says an out-of-grammar utterance (by not specifying the state), which correctly gets rejected, but the error prompt isn't helpful. Instead of asking for a city and state, it asks if she wants Fulton, California.

Since users can either say yes or no, or name another city and state, what should we do if an utterance is rejected? The user might have been trying to say yes or no, in which case an appropriate error prompt would be "Sorry, I didn't understand. If you want Fulton, California, say 'yes'. Otherwise, say 'no'." Or, the user might have been trying to change the city, in which case an appropriate prompt would be "Sorry, I didn't understand. Please say a city and state, like Boston, Massachusetts."

Fortunately, if an utterance is rejected in this yes/no state, we can make an educated guess about what the user wanted to do. Yes/no grammars have low rejection rates, while city-state grammars are significantly more complex and have higher rejection rates (available data shows rates as much as four times higher). So, if an utterance is rejected, it is likely to be because the user named a city and state.

Based on this reasoning, when a reject error occurs, we can assume that the user doesn't want to stay in the yes/no state, and we should ask them for a city and state:

System: Getting movies for Fulton, California. Is that right?
User: Fremont.
System: Sorry, I didn't understand. Please name a city and state, like Boston, Massachusetts.
User: Fremont, California.

When the user's utterance gets rejected, the system compensates by guessing that she is trying to name a city and state. It takes the reject as an implicit 'no', and takes her to the basic city-state question. This type of approach enables the creation of efficient, effective error handling stemming from a thoughtful, data-driven design approach.

Avoiding Endless Loops
Even with all the above improvements in place, though, errors are still bound to happen. The important thing is that the errors don't stop the user from accomplishing their task. One pitfall that users can fall into is an endless loop of confirmations. In the following example, the user is having trouble getting a city recognized, because it isn't actually in the grammar:

System: What city and state are you in?
User: Rockridge, California.
System: Rohnert Park, California. Is that right?
User: No.
System: Sorry about that. Please say your city and state again, like 'San Francisco, California'.
User: Rockridge, California.
System: Rockcreek, California. Is that right?
User: No.
System: Oh, sorry. Please say your city and state again, like 'San Francisco, California'.
User: Rockridge, California.
System: Rollings, Pennsylvania. Is that right?
User: No.

If the user is persistent, they could keep saying 'Rockridge, California' and then say 'no' to the confirmation all day, and never get anywhere. To limit such unproductive behavior, the system should keep track of the number of errors, and switch to another strategy after some threshold has been reached. This alternative strategy might consist of capturing the information in another way (such as by zip code), transferring to an agent, or skipping the step if the information is optional.

When counting the number of errors, it is crucial to include various different types of behavior, including no match errors, no input errors, saying 'no' in confirmations (as in the example above), and saying 'go back' in confirmations. In some cases, the system should also count other indications of trouble, such as asking for help or requesting to speak with an agent.

By keeping track of the total error count and providing an appropriate fallback mechanism, we can help keep errors from spiraling out of control. With these changes, the city-state dialog would look this:

System: What city and state are you in?
User: Rockridge, California.
System: Rohnert Park, California. Is that right?
User: No.
System: Sorry about that. Please say your city and state again, like 'San Francisco, California'.
User: [No speech]
System: Sorry, I didn't hear. Please say your city and state, like 'San Francisco, California'.
User: Rockridge.
System: Sorry, I didn't understand. Let's try this another way. What's your zip code?
User: 94618.
System: Got it, 94618.

In this example, there are three total errors: first, the user says 'no' in the confirmation, then has a no input error, then a no match error. The threshold for total errors is set to three, here, so after the third error we switch to the fallback strategy of capturing the zip code instead of the city and state.

Confidence-Based Confirmations
Explicit yes/no confirmations are a good way to make sure the system gets the right information, but they do add significantly to the time it takes to complete a task, and can be annoying when users have to get through many of them in a row. To speed things up, we can use a confidence-Based Confirmation. After the user speaks an utterance (such as "Fremont, California") and the recognizer has made a recognition hypothesis, the system checks the recognition confidence, and does different behavior based on the result.

If the confidence is very low (below the standard rejection threshold), the system probably did not recognize the utterance correctly, so the system will play a no match error as usual, like "Sorry, I didn't understand. Please say your city and state again".

If the confidence is medium (above the standard rejection threshold but below some other threshold), there is a reasonable chance that the system recognized the utterance correctly, but it's not very certain. So, the system will do an explicit yes/no confirmation to be sure. That yes/no confirmation is the same one we have been describing all along.

However, if the confidence is high (above this other threshold), it is likely that the utterance was recognized correctly. Since we're already pretty sure we got the right result, we don't need to do a yes/no confirmation. Instead, we can save time with an alternative flow, such as just echoing back the recognized item, or not echoing it back at all.

As an example, let's compare the behavior for a medium-confidence recognition and a high-confidence recognition, in the main menu state of an application :

Medium-Confidence Behavior:
System: How can I help you? You can say 'account balance', 'make a payment', or 'find a store'.
User: Um, yeah, I'd like my account balance.
System: That was account balance, right?
User: Yes.
System: You currently have $24.95 in your account… [etc.]

High-Confidence Behavior:
System: How can I help you? You can say 'account balance', 'make a payment', or 'find a store'.
User: Account balance.
System: You currently have $24.95 in your account… [etc.]

In the medium-confidence case, the system uses a yes/no confirmation, while in the high-confidence case, the system skips the confirmation and goes directly to read out the account balance.

Conclusion
Yes/no questions provide an example of a state that seems simple to design, but a thoughtful design improves the application greatly. A well-designed system accounts for the context in which the question is asked, the caller behavior observed in the deployed system, and the variety of error scenarios that callers may encounter. By changing the approach to grammar coverage, prompting and error handling, and by avoiding endless loops and using confidence-based confirmations, we shortened the task completion time, and reduced out-of-grammar rates. The result is a far better user experience, which translates into higher rates of task completion.


About Dr. Lisa Guerra:
As Vice President of Engineering, Lisa is responsible for the development of BeVocal's application products as well as customized application solutions to meet the business objectives of BeVocal's customers. Lisa is an experienced product development manager responsible for leading all of BeVocal's application development initiatives since the company's inception.

About Dr. Ryan Bush:
As Senior User Interface Design Lead, Ryan is responsible for the user interface design of BeVocal's applications. Ryan has designed many of BeVocal's applications, both in English and Spanish, and leads the other members of the design team. His background in Linguistics gives him a passion for natural and intuitive interfaces.

About BeVocal:
BeVocal provides managed call automation solutions which help enterprises and service providers reduce operating costs while significantly improving customer service levels. Working with customizable, packaged voice application modules optimized for performance and usability, BeVocal's professional services team can deliver high quality solutions deployed on BeVocal's hosting network to answer, automate and route customer care calls. BeVocal also provides a carrier grade application hosting environment, and tools for customers and partners to build their own voice applications using BeVocal Cafe, the industry's top VoiceXML development environment.

Today's Tip of the Day - Keep Cost In Perspective

Read today's tip or listen to it on podcast.

Published: Friday, February 4, 2005

Printer Friendly Version Printer friendly version

2024 Buyers Guide Automated Call Distributors

 
1.) 
Premium Listing
Call Center Studio

Call Center Studio
Call Center Studio is the world’s first call center built on Google and is one of the most secure and stable systems with some of the industry’s best reporting. It is one of the most full-featured enterprise grade systems (with the most calling features, one of the best call distribution, outbound dialing features and integrations—including IVR, AI Speech Recognition, blended inbound/outbound calling and includes Google’s new Dialogflow and Speech API. Call Center Studio is the absolute easiest to use (with a 10 minute setup), and is the price performance leader with lower equipment cost and less setup time.


2.) 
Teckinfo Solutions Pvt. Ltd.

InterDialog UCCS
InterDialog UCCS inbound call center software caters to all incoming customer requests. These incoming requests can come through any channel of customer’s choice e.g. voice, video, email, chat, WhatsApp, facebook etc.. company page or from an integrated website chat. Using InterDailog UCCS call centers can respond to inquiries of the customers and they can also register the complaints of customer as a customer support desk.
 

About us - in 60 seconds!

Industry Champion Award Leaderboard

Most active award entrants in the past 48 hours! - Vote for Others / About Program
Submit Event

Upcoming Events

The 19th AMERICAS Annual Best Practices Conferences are here! Meeting Point for the World's Best Contact Center & CX Companies Read More...
 31813 
Showing 1 - 1 of 3 items

Newsletter Registration

Please check to agree to be placed on the eNewsletter mailing list.
both ids empty
session userid =
session UserTempID =
session adminlevel =
session blnTempHelpChatShow =
CMS =
session cookie set = True
session page-view-total = 1
session page-view-total = 1
applicaiton blnAwardsClosed =
session blnCompletedAwardInterestPopup =
session blnCheckNewsletterInterestPopup =
session blnCompletedNewsletterInterestPopup =