Sramana Mitra: Is there anything else that is significantly different in the text analytics field currently?
Rick Kieser: The other element is taking a look at foreign languages through text analytics and how that evolution is taking place. The context of generating sentiment from English vs. French vs. German is very different in terms of creating rules around it. Within our platform, our base technology can handle any character-based language – any foreign language. It is very simple to be able to operate in this platform. On our NLP engine, we have 33 languages on sentiment. It is not a trivial process in terms of being able to glean sentiment out of different languages. That is one of the areas we are constantly working on for our international customers, increasing usability in a multilingual environment.
SM: What about speech? What is your impression on the state of the union in the speech processing industry?
RK: Within our base product, we have automated translation capabilities. They are to be used by a trained linguist. From a market research perspective, everything that gets coded and classified has to be very precise. It can’t have a lot of “line loss.” We have looked at a number of different voice-to-text technologies, and we have not found them robust enough to put into our platform because of the difficulties in having multiple speakers, for example. When you have multiple speakers in a call center or if the recording is poor, it makes it incredibly difficult to handle. That is an area where I see there are going to be improvements in terms of being able to make this [technology] more robust. Right now we leverage a lot of transcription capabilities within our own firm to be able to do it, but we are not using voice-to-text on a wide scale; we are experimenting with it.
SM: Are there any open problems in related areas that need to be worked on and that you would encourage entrepreneurs to consider?
RK: Since we use a hybrid approach, the multiple application of different technologies around unstructured text is really the issue. I want to use the analogy of a single speed bike vs. a three-speed bike vs. a road bike. If you are going to be in the Tour de France, you are not going to get on a single-speed bike. Today many of these NLP competitors just have a single-speed bike. We think we are best in class, because we have three different gears we can play upon. But there are innovations that are making these things easier to use and integrate. Today, if best in class is being able to blend three of these in a seamless manner, I think in the next five years this blending will be much more intelligent, where you will potentially even have optical recognition or just traditional text messages and then come in there and be able to tell what this is.
It is a really exciting field. It has taken a long time to develop. If you look at the customer experience management (CEM) industry, a decade ago they were nonexistent. People didn’t know those terms. Now CEM is a $2.6 billion industry and is growing 20% per year. It is exciting to see this. Blending these, making them more robust, and making them able to handle all these different channels of feedback is very exciting to see, and we are happy to be a part of it.
SM: Thank you, Rick.
RK: Thank you.