Wednesday, April 8, 2009

I don't think I said what you think I said - Interviews and "Natural Language"

"Hi, I've just got a few questions." Not everyone likes it, but I will admit I do. I like to get interviewed. I like to be asked questions. I like to feel like someone cares about my answers, relative truth aside. I like to think the thoughts and words will make their way into a tight, punchy piece that makes a reader or a thousand think a different thought or care more. But, I know enough from a media course and life to know better than to expect much of that.

Now, this week, I received several mentions in a Speech Technology magazine article and I'm glad. It's a bit ironic, given news I'll get to in another post, but I'm enjoying it. At the risk of not getting interviewed in the future, I do need, though, to straighten out a tiny thing or two.

So, if you're not a speech geek, or wanna-be speech geek, skip to the next post, otherwise, a couple of discussion points:

First, the article is a great intro to current thinking about the use of open-prompting (soliciting content constrained by context, not wording). I favor this approach when it makes sense and can be delivered properly. In general, my thoughts are represented well. But, to get to the point, I didn't actually assert "that callers shouldn’t be exposed to a hierarchy of more than five categories." I do think menus structured like that can be problematic and are frequently done poorly, but research (Hura & McKienzie) and deployments (McKienzie, Levine) have shown that the right combination of wording and delivery can allow menus to be fairly lengthy and still be effective. I agree with those findings.

Moving on, the article discusses the idea that a high-frequency example is very important, which is correct. However, open-prompts should almost never start with "How may I help you?" Not only is there a register problem, but putting the call to action before giving the caller space to respond is generally a recipe for disaster, despite enabling barge-in.

Next, "performance anxiety," my (unattributed) quote, is not a technical term. Funny? Yes. Not technical.

Then, the sentence reading "Users are given the option of accessing them by saying something like What are some choices?" really should read "Callers can be given the option to access them by suggesting they say something like "Give me some choices." The difference appears subtle, but trust me, it's important. Control and certainty are increased by this very much directive statement, whereas the other feels too much like browsing.

Lastly, I need to put a finer point on "it’s important to design the back-off menu open-ended like the first prompt." Actually, the true matter is to allow not a full range of open responses, but to look for responses in tuning data for the menu and adding to the grammar utterances that are clearly a response to the open prompt but not in-grammar for the menu. For example, a caller saying "my internet account" to a menu asking them to specify whether their call is about their bill, an order, tech support, or an appointment. And of course adding logic to handle such utterances. Doing this will be effective and caller-pleasing without the trouble of a parallel SLM grammar.

So, these are not slams, but rather just trying to make sure that the right information is out there. The article is, I think and hope, a good discussion starter. Including the last few paragraphs, which are sort of not-directly-on-topic, but are vitally important nonetheless.

Please add your take on the article here. If you want to know more about the concepts presented there or here, or you want to interview me ;), please let me know: phillip at phillipwhunter dot com.


UPDATE: Eric Barkin responded very graciously to my post on his SpeechTech blog page. I appreciate his comments and especially agree with the points about the need for a larger design improvement discussion.