DDJ: We're pleased to be joined by Luca Scagliarini, vice president of strategy and business development for Expert System, a company that specializes in semantic intelligence. So Luca, what is "semantic intelligence"?
LS: Semantic intelligence is a technology and approach to natural-language processing and unstructured information management in general. At Expert System, we've embedded in our software the technology to analyze text at four different levels, resulting in an understanding of the author's intended context for readers.
The four approaches include:
- Morphological Analysis.The capability to understand word forms; for instance, to recognize that the words "dog", "dogs", and "dog-catcher" are closely related.
- Grammatical Analysis. The capability to understand the parts of speech; for instance, "There are 40 rows in the table" uses "rows" as a noun, versus "She rows 5 times a week" uses "rows" as a verb.
- Logical Analysis. The capability to understand how words relate to other words; for instance, "Jeffrey Skilling, represented by Attorney Daniel Petrocelli, is married to Rebecca Carter". Rebecca is married to Jeffrey, not Daniel.
- Semantic Analysis (disambiguation). The capability to understand the context of key words; for example, "I used beef broth for my soup stock" uses "stock" in the context of food versus "The company keeps lots of stock on hand" uses "stock" in the context of inventory. Semantic Analysis looks at surrounding words as clues and a 3.5 million entry database of definitions and relationships to understand the context.
Many current technologies rely on approaches Morphological and Grammatical Analysis. At best, they use statistical tricks to attempt Logical Analysis, but no one except us has tackled Semantic Analysis. We put all four approaches into one tool. In fact, the entire approach is exactly as we all learned as children and still use each of the above approaches in decoding what we read everyday.
DDJ: Why is semantic intelligence referred to as "the enabler of mobile search" and not search in general?
LS: The advantage of this approach is that it translates into higher precision and recall -- for any kind of search. It increases precision because it enables the possibility to search by concept (that is, to retrieve only documents where the keyword "stock" is used with the meaning of broth and not the documents where "stock" is used with the meaning of "shares") and it increases recall because it lets users, if they require it, to simply let the system search for "car" and automatically extract all the kinds of cars because it understands the meaning of the word.
The capability to understand language also enables answering to natural-language based questions. For example, if I want to know who sang "Imagine" and I have access to a repository where this answer is present (Wikipedia, for instance) I can ask the system the question in natural language and the system returns the exact answer.
Due to the specific of mobile search -- small screen and limited capability to go through long lists of links -- this feature is the enabler of this kind of search.
DDJ: Is there a web site that readers can go to find out more information on semnatic intelligence?
LS: Yes, they can go to www.expertsystem.net.
