More mature, less awkward teen, and zero used car sales
I just finished editing all of the Care abstracts going into this year's Conversations program. I'm quite pleased with the lineup that we've scheduled -- everything from in-depth technical sessions on robust parsing and SpeakFreely deployments to higher level discussions of how to develop a speech strategy. I only wish I could be in all 32+ sessions at once, because I know I have a lot to glean from all of these!
One thing that I have noticed is that the way we talk about speech recognition in general has matured significantly -- certainly over the last decade, and especially the last few years. I guess we're out of that awkward teenager stage where everything is just so... *dramatic.* (Fewer exclamation points and "it's fantastically amazing!" pitches.) Still, I went through and personally zapped several words that have always bugged me when it comes to describing speech systems... they're lazy used-car-sales selling words, IMNSHO. They had infested a few abstracts and I felt it was my sworn duty as an engineer-turned-marketer to purge them from our speech vocabulary:
Simply. If I had a nickel for every time someone said you could do something "simply by speaking"... the truth is, it's hard to resist the allure of making the argument that we've been speaking since we were toddlers, so it's *gotta* be easier than typing/touchtone/menu navigation/what-have-you. But anyone who's gotten close to a speech system knows that there's nothing simple about speaking. There are out-of-grammars. Actual misrecognitions. Prompts to design (and redesign when unanticipated problems arise from ill-informed callers.) As any designer or user of a speech system knows, there are pitfalls aplenty. Speech has enough positive benefits and advantages over other interface options that we don't need to overpromise and underdeliver by calling it "simple."
Conversational speech. You'll also notice in this year's abstracts almost no mention of conversational speech from Nuance personnel. Yes, our goal is to make every speech interaction resemble a conversation, but it's a predictable conversation with rules and structure. True conversational speech involves lots of asides, excuse-me's, and maybe some noncommittal grunts for good measure. I personally think that the moniker "conversational speech" also invites uncomfortable expectations.
Finally, I would also like to say with pride that in these abstracts, nothing is ever leveraged, very few technologies are utilized, no technologies, products, or solutions are ever described as new and/or improved, and the few sessions that talk about best practices have the credibility to do so. Thank me later!
I wish I could also say that I had removed all references to the term natural language, but sadly, I've found it a necessary evil in the speech world right now, with some qualifiers. The problem is that saying you're going to talk about natural language is like saying you work with computers. It's a term that's so broad and enveloping that it runs the risk of becoming meaningless. First of all, what qualifies something as "natural language" -- is it having enough synonymous responses in the grammar to fake it? Is it having a full linguistic parser breaking down what someone is saying and ascribing it meaning? Is it being 'good enough' to guess at what someone is saying? Is it ignoring the filler words to make a decision about the next prompt? Once you get past that, there's the question of whether "natural" is a good thing! Most people tend to think so, until a caller is so natural that the system can't handle it... then of course the caller's to blame. There are a large number of sessions about natural language, and what I'll say here is that we've tried hard to make those sessions about a specific implementation of natural language, like robust parsing, or SpeakFreely, or call steering, so that you can determine how opening up what a caller can say may be used to your advantage.
Comment Notification
If you would like to receive an email when updates are made to this post, please register here
Subscribe to this post's comments using