Trio of Tech Firms Go Natural with a More Human Interface

Vocalcom, Umanify, and Loquendo are combining their technological know-how to create Saynergy, a natural language-based digital assistant, the companies announced recently.

The joint venture is said to provide users a more lifelike human-to-machine interaction with two base interfaces: an open-ended voice interface and an open-ended, Web-based, written interface.

Each firm involved brings a special skill set to bear on the product. Vocalcom provides the dialogue model for the solution; Umanify some reactive human avatars for graphical implementations (thus far limited to the Web and mobile Web); and Loquendo the text-to-speech (TTS) and speech recognition capabilities.

The joint venture evolved out of a text-based platform developed by Natural Language, a part of the Vocalcom Group, explains José Quesada, technical director at Natural Language. The dialogue models developed in this project power the speech functionalities. However, even as Natural Language was working on text specifically, it had its eyes set on a larger goal.

From its inception, the Saynergy project was aimed at developing a system that could more organically understand human communication on a semantic level, regardless of form. That is to say, the aim of the system was not to recognize individual words, but user intent.

Quesada sees the ability to incorporate speech recognition into this system as akin to giving a brain that already understands language ears. This, however, isn’t to say that Loquendo’s participation in the project is superficial. Loquendo's TTS and speech recognition are like an actual biological mouth and ears, fundamental parts of the organism.

The two companies spent an entire year exploring how their technologies could be integrated, making sure they could work cooperatively at a deep and fundamental level.

“The main goal of this technology is the understanding of natural language, [that is] people speaking or writing,” Quesada says.

According to Quesada, the results of those efforts have borne significant fruits. Before releasing the solution, a case study was initiated with a third-party enterprise. The study sampled as many as 20,000 real-world calls where users had specific goals they were trying to negotiate with the solution. At its end, Quesada boasts that there was between 80 percent and 85 percent semantic accuracy of the calls; semantic being the operative phrase.

As he explains, “In order to evaluate our approach, it’s not a good idea to use the classical model of number of words currently recognized or not. Instead, our idea is to evaluate the semantic accuracy of the process. If a user wants to fly to London tomorrow, perhaps, and says I would like to fly tomorrow to London, [in the classical approach] you measure the number of words the system recognized. You obtain an interesting figure, but the most important part is: Have you been able to understand the goal of the user?”

It is this understanding that is measured in the semantic accuracy figures, Quesada says.

So far, Saynergy has only been implemented in Spain, where Vocalcom has the highest concentration of offices. It is, however, expected to roll out other implementations through the company’s other distribution channels and offices, located in 38 countries.

“Vocalcom has initiated a very strong marketing strategy for natural language,” Quesada says. “In the last few months we have made several presentations in France, Belgium, Germany, and Poland. We’re going to present in two weeks in the Berlin Call Center World 2009...[where we’ll] present an application in four languages: English, French, German, and Spanish.”

SpeechTek Covers
for qualified subscribers
Subscribe Now Current Issue Past Issues