The Taxonomic Underpinnings of Cognitive Computing

Although it’s making a resurgence of sorts with the concepts of neuro-symbolic AI and what Gartner has termed composite AI, machine reasoning has traditionally been overlooked in recent years. Rules, axioms, inferences, and logic-based systems are integral aspects of cognitive computing, particularly as it underlies machine intelligence.

Taxonomies are the substrate of nearly each of these aspects of symbolic reasoning, which oftentimes involves knowledge graphs. The standardized vocabularies, definitions, and hierarchies of definitions taxonomies encompass usually directly correlate to the business concepts upon which cognitive computing systems reason.

Additionally, the breadth of use cases taxonomies support for cognitive systems includes mainstays such as enterprise search, text analytics, and linked data deployments, in which different business units—or organizations—are able to seamlessly exchange information with one another for sophisticated interoperability for smart cities, ambient computing, and cross-regional transportation.

In these applications and others “you can fine-tune the abstraction level of your search by using taxonomies,” Franz CEO Jans Aasman indicated. “Those taxonomies also provide you with all the synonyms of what you’re looking for.”

Cognitive search is one of the most universally applicable cognitive computing functions taxonomies directly enable. As Aasman mentioned, they not only do so by furnishing synonyms of various vocabulary terms throughout the enterprise, but also by assisting with semantic search. The efficacy of semantic search, of course, vastly exceeds that of traditional keyword search and is predicated on understanding the intention of what users are looking for—to a certain extent, regardless of how that intent is expressed.

The cardinal advantage of taxonomies is they incorporate defined vocabularies into what Aasman called “hierarchies of meaning.” For example, a networking company has specific names of products, some of which might be routers, which are examples of hardware. Each of those concepts (product names, routers, and hardware) is a level in a hierarchy of machine understanding. “So, if you ask the database to give me every document that talks about routers, then it will go and look in the taxonomy,” Aasman explained. “It will say, what kind of routers do I have; these are the routers; these are the names that I use for them, and these are the alternative names. Now, let me search through all the documents to find them.”

Text Analytics

The blueprint Aasman described of standardizing enterprise terminology and definitions, creating hierarchies of those definitions, and using them for cognitive search, is also the basis for text analytics and non-statistical Natural Language Processing. For text analytics, one of the first steps is users simply input their various taxonomies into an entity extractor. For the router use case above, that extractor “will extract every router it finds in any text and it puts it as triples,” Aasman commented. “So, now you can do a SPARQL query and say, ‘find all the documents that have a router,’ and it will be instantaneous because you took out the entities.”

This approach to text analytics is considerably assisted by leveraging vocabularies as the basis for designing business rules or axioms for understanding natural language text. Once speech is converted to text, this same approach is also helpful for speech recognition. “You can use taxonomies to train speech recognizers,” Aasman pointed out. Doing so provides more nuanced, granular understanding of speech to text deployments that oftentimes evade the discernment of pure machine learning model approaches, particularly for uncommon terms like names, products, and proprietary information.


Lastly, taxonomies are critical for facilitating interoperable understanding of terms, definitions, and business concepts within and between different business units or firms. For example, there is an almost countless array of taxonomies within the healthcare industry for different areas of specialization, medical procedures, and other points of distinction between providers, their countries, and more. However, there’s now an overarching taxonomy that tames this complexity and these points of differentiation.

“It’s called UMLS,” Aasman mentioned. “It’s a system that harmonizes all the existing taxonomies in healthcare.” The implications of this use of taxonomies are considerable. Taxonomies consisting of a plethora of other taxonomies are integral for sharing information between them, helping to pave the way for systematic interoperability—particularly with the machine understandable data that taxonomies predicated on semantic standards involve.

In Retrospect

There is much more to cognitive computing than statistical approaches and machine learning. Taxonomies are a critical aspect of intelligent reasoning systems that assist with interoperability between different systems, text analytics, and cognitive search. Utilizing this form of cognitive computing is imperative for truly creating machine intelligence.       

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.

Opinions expressed by contributors are their own.

About Jelani Harper

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.

View all posts by Jelani Harper →