The universal applicability of enterprise knowledge—across use cases, domains, and languages—is widely understood. It’s foundational to the history of Artificial Intelligence.
And, it’s likely the main reason adoption rates for knowledge graphs have steadily inclined of late, making them one of the most utilitarian forms of AI available today.
True knowledge graphs are extensible and predicated on standards designed to share data of any type. Such graphs are inherently composable, enabling users to either combine them or enrich them with knowledge of all sorts.
These options are critical for not only simplifying the management of enterprise knowledge for Natural Language Understanding deployments, but also for redoubling the value organizations reap from knowledge graphs across a burgeoning array of use cases.
“More knowledge always wins against less knowledge,” noted expert.ai CTO Marco Varone. “I believe that is true in general, not only in our space. It’s always better, when you need to solve a problem, to get a bit more knowledge than less knowledge.”
Organizations can extend semantic knowledge graphs across languages and business domains to get much more knowledge about complex language understanding applications, drastically improving this technology’s ROI.
Across Language Barriers
Top language understanding solutions involve knowledge graphs predicated on semantic standards that are extensible across languages. This approach is based on an ontological foundation in which the world’s concepts are modeled. Then, there’s an additional component in which those concepts are applied to different languages via constructs such as taxonomies, thesauri, and the like. According to Varone, “This multi-language knowledge graph is something that is really generating a lot of value. Many of our customers are multinational companies. We have one customer using 12 languages at a time. Many of them are using two or three: the local language and English.”
Although there are boundless monetization opportunities for implementing NLU across languages, an e-commerce use case succinctly demonstrate their worth. Regardless of what a company’s native language is, “With e-commerce, everybody can sell anywhere in the world,” Varone mentioned. “And what do you use to do this? English.” Thus, firms can empower chatbots with AI for natural language interactions with e-commerce customers, both parochially in their native language and internationally in English. From a technological perspective, however, there’s just one knowledge graph—regardless of how many languages to which it’s applied. “The more the two [linguistic applications] are alike, the easier it is to implement,” Varone observed. This approach reduces costs (which would otherwise increase to build multiple knowledge graphs), while compounding the value from the initial knowledge graph investment.
The horizontal domain use of knowledge graphs for language understanding is just as convincing as the use of these graphs for different languages. Instead of having different knowledge graphs for respective domains such as marketing, sales, or other business areas, organizations can simply employ one graph to encompass these or any other domains. “It’s a multi-domain knowledge graph, but when you apply it to specific domains, it’s common that you need to enrich it,” Varone explained. The knowledge enrichment process can take many forms, including additions of new glossaries, taxonomies, and nomenclature. Organizations can get this information from experts, or have experts oversee certain machine learning techniques, to add this knowledge to their various domains.
“During the years we have implemented what we call an extension to the knowledge graph which is…an enriched ontology and enriched thesaurus for the domain,” Varone indicated. “We give the possibility to our customers to add the new knowledge for a domain themselves.” This knowledge enrichment phase is best handled by domain experts well versed in the knowledge of a specific business function—as well as in how to properly enrich knowledge graphs for this purpose. “They need a bit of training, for sure,” Varone acknowledged. “But it is not super complex. I must say that it is not something that you can do from day one. You need a bit of experience and training, because knowledge is a complex element to manage.”
Whether companies want to employ a single knowledge graph for language understanding use cases across languages or domains, both options increase the amount of knowledge at their disposal. This additive knowledge quickly becomes equally additive for the value gained from it. “Let me make this example,” Varone said. “Do you prefer to have a surgery on you by a surgeon who has just finished school, or one that has 20 years of knowledge and experience? Everybody will tell you they would go with the person with 20 years of experience. The same is true in our field. It’s better to have as much knowledge as possible.”
When that knowledge is on a single knowledge graph and applied to different use cases and languages, organizations not only democratize it, but also increase its ability to solve business problems.
Featured Image: NeedPix
Jelani Harper is an editorial consultant servicing the information technology market. He specializes in data-driven applications focused on semantic technologies, data governance, and analytics.