Amtera, a Brazilian-based startup whose main solutions are focused on semantic enterprise search and knowledge management, is exploring the API business model.  Its first step in that space is its new semantic relatedness API, which is a piece of its overall pipeline technology for querying Linked Data. Its main products are Esprit for enabling semantic search by meaning rather than keyword, and Sentient, for social media analysis.

The API, the company says, supports applications accessing large-scale distributional knowledge bases, offering word sense disambiguation and classification so developers can deliver knowledge models assured of the correct determination of words.

The API, as it is currently exposed, is for comparing pairs of terms in text, and for disambiguating the words into the right meaning with a minimum level of information. “Star,” for example, can refer in an ontology or taxonomy to a celestial body or to a performer or actor. The word can be disambiguated into its proper sense via its pairing with other information found in surrounding text – if the context for the use of the word “star” involves the term “Angelina Jolie,” the pairing is with the latter meaning of star.

The API’s core operation is to compute the semantic relatedness measure, to allow the systematic determination of the semantic proximity between terms. “This way allows a more flexible way to align ontologies, to compare things in a more flexible and complete way,” says co-founder, partner and CEO João Gabriel Oliveira. “When you create knowledge models based just on ontologies, though they might be extremely correct, they tend to be not that complete. You couldn’t cover all the possibilities. We think eventually, by using both together, you can cope with this gap in completeness.”

The company’s API plans, Oliveira says, involve scaling elastically through the cloud. “Right now,” he says, “we could compute about 300 pairs per second, which is very much….I don’t know any other implementation of this kind of solution, this kind of semantic model, that is that fast.” Because it is a linearly scalable model, Amtera could also increase the throughput, he says.  “The technology itself is highly optimized, so we can elastically increase the scale and easily cope with performance requirements,” he says.

Currently, the API service is available in Portuguese and English versions, but once the company has pinned down its method of validating the model, it can be easily generated in other languages. Expect Spanish, French and German capabilities soon. There is a freemium offering that supports up to 5,000 pairs per day, as well as licensed options.

Oliveria notes that the company also could exploit other aspects of its core technology in the API format, such as relation extraction.