Intellego: Technology that Understands You

Promptu's groundbreaking Intellego spoken language understanding platform is based on two components: open dictation automatic speech recognition (ASR) to convert the user's natural speech into text, and natural language understanding (NLU) to extract the text's meaning.

By combining these two technologies, available via the cloud or fully embedded, Promptu’s Intellego platform makes voice-powered applications flexible, accurate, and useful. Working with Promptu, customers can quickly adapt the technology to new designs and features, implement new use cases and behaviors, or tune systems to improve the user experience.

Automatic Speech Recognition (ASR)

The key to Promptu’s Intellego ASR is open dictation. Users are no longer forced to memorize—or more typically, haplessly try to guess—the command phrases embedded in a fixed, rigid grammar. With Intellego, any meaningful sequence of words can be recognized. Users can speak their queries or commands in natural fluent language, such as “I'm in the mood for Italian food,” or “Find a good Italian place near me.” Read more

Promptu has developed proprietary methods for rapid adaption of its ASR technology to specific application domains, vehicles and devices. The vocabulary and language model are tailored to the words and phrases typical of the application domain, such as business search, command and control, or audio and video content selection. Likewise the acoustic model can be customized to audio as captured by a specific microphone, colored by a given codec, or corrupted by the typical road noise spectrum of a particular automobile. Collapse

Natural Language Understanding (NLU)

A generation ahead of the competition, Promptu’s Intellego NLU processes its inputs as a whole, understanding each word or name that appears in a sentence in the context in which it appears. While a human knows the difference between a search for “movies with Robert Redford” and “movies by Robert Redford,” common keyword-based or text-matching search tools do not. The same goes for queries about “how to shoot a video” versus “how to watch a video,” and the distinction between a business that “sells nails” and one that “does nails.” Intellego knows what the user means, with high precision and high recall. Read more

Promptu's Intellego NLU technology uses rules, not statistical models, making it easier to fine-tune system responses and implement new use cases, which is perfect for iterative design strategies. The semantic network is pre-populated with over 300,000 word lemmas and collocations, and includes a preprocessing step that performs part-of-speech labeling, lemmatization, and word sense disambiguation.

While the semantic network is large enough to cover most application domains, Promptu can work with system designers to expand the network with pertinent words, concepts, and named entities if necessary.

These rules and the preprocessing step lower the workload for designers by accounting for synonymous words and phrases. It also lowers response times and operating costs. If an input has multiple meanings, the technology assigns a confidence score to each interpretation and asks the user which is correct—just as a human does when unsure how to interpret a phrase.

Promptu's NLU technology makes voice-based systems less prone to error. Even if an input doesn't exactly match what the designer anticipated, the system can extract the correct meaning. This ability is key to dealing with the diverse ways in which people express themselves, but it also compensates for minor speech recognition errors. As long as a few essential words of the user’s request are correctly transcribed, the system can discern the user’s intention, and respond appropriately. Collapse