Getting a computer to understand English is a huge challenge for computer scientists. Getting one to understand 15 different languages seems monumental. That is, however, exactly what LinguaSys, a provider of natural language processing software, has done. Curious, TMCnet recently caught up with Brian Garr, CEO of LinguaSys, at SpeechTek in New York City to discuss the company’s recent efforts in Natural Language Understanding and text analytics.
Traditionally, Garr said, natural language processing is done using statistical language models, where scientists build vast, complex models mapping the various ways in which a word might be used. Spoken or written word is then compared against this data to determine what is being said, based on probability.
LinguaSys, on the other hand, strayed from that model. Instead, it leverages a linguistic engine called Carabao, which is able to analyze and parse concepts. That is, it is able to understand object relationships between words, creating a semantic web of parent and child terms. For example, it understands that “food” is the parent object of “pizza,” under which “topping” and “size” may be children. Using these conceptual webs, known as “sequences,” LinguaSys is able to determine context regardless of language structure or morphology.
According to LinguaSys, these sequences “allow looking up patterns according to any combination of attributes, be it a grammatical feature (e.g. part of speech, morphological case, inflection pattern), a stylistic tag (News - Alert) (e.g. regional usage, medium where it is typically used, sentiment), a reference to a concept (e.g. elevator will capture alsolift), or a whole category of concepts e.g. (vehicle is a category which includes car).”
Being able to search for and quantify exactly what is being said about your company, say, for example, on social media, is a logical enterprise application of this technology. Indeed, LinguaSys has recently announced its Sentiment@Work language processing solution, designed to harvest actionable intelligence from social media, online content, news media, consumer sites, e-mail and big data.
“If your company has to know what global customers are saying in their own languages, even when they offer opinions about multiple subjects in the same Tweet or content unit, Sentiment@Work is the solution,” said Garr. “For example, our clients want to know what a Russian-speaking guest really said – in her own language - about both the hotel room and breakfast in the same Tweet.”
The company hopes to help small and medium sized business compete in the global marketplace by offering the service on demand, on a subscription basis. Users can pay by the day to receive granular analysis of multi-lingual content sentiment, down to the individual clause level.
With this insight, LinguaSys says, companies can be more in-touch with prevailing customer sentiment, allowing them to take both appropriate and timely business measures.
Edited by Alisen Downey