The Inbenta chatbot uses natural language processing for their chatbot to provide enterprise-level service. Inbenta CEO, Jordi Torras (JT), granted me an exclusive interview to discuss how the company developed the Inbenta chatbot.

Natural language processing doesn’t get discussed much outside of niche communities. But this practice informs the future of chatbot technology at-large.

I recently wrote about the differences between chatbots and virtual assistants. Though I covered the difference, none of them employed natural language processing to the same degree that the Inbenta chatbot does.

How does natural language processing inform this chatbot and how can it help you?

Identifying the Problem They Can Solve

Jordi Torras spoke with great fervor, showing his passion for AI and technology. He believes that, though the possibility of job risk due to AI is scary, it is not as terrifying a problem. AI is meant to help improve human lives — not replace humans.

Lucky for us, the latest study about jobs at risk of automation said it was only around 14%.

But Torras needed to identify the best way for his company to leverage AI. After all, there are almost limitless possibilities and applications for neural network backed technology. This search led him to chatbots, natural language processing, and enterprise search.

“Search is a big space in computing With two kinds of problems: internet search (which is solved) and enterprise search. There is no equivalent to Google in that space.”

Founded in 2005, the company now supports 24 languages including:

  • English
  • Spanish
  • German
  • French
  • Russian
  • Japanese
  • Chinese

They also serve some significantly high profile clients such as Groupon, Ticketmaster, Change.org, and Schlage.

Besides their chatbot product, Inbenta also offers content management, long tail SEO help, email management, and other services.

But how does natural language processing play a role in their chatbot?

More importantly, how did Inbenta resolve the two key questions revolving around chatbot implementation and interaction: relevancy and disambiguation?

image of Jordi Torras from Inbenta for article Inbenta Uses Natural Language Processing for AI-minded Chatbots
Jordi Torras | Inbenta

Discerning Intent for Relevancy and Disambiguation

One of the first problems Inbenta set out to solve: disambiguation. Discerning the intent and meaning behind the words leads to more intuitive chatbot experiences.

Torras used the example of the word “like” to describe how intent can shift.

Inbenta treats the preposition “like” as a totally different, unique word from the verb “to like”. As such, they both get different IDs in the Inbenta system.

“IF I SAY ‘IT’S LIKE YOU DON’T LIKE ME’, I SAID ‘LIKE’ TWICE, BUT EACH MEANING WAS DIFFERENT. THE FIRST ‘LIKE’ IS A PREPOSITION AND THE SECOND ONE IS A VERB.”

This thought process extends to all words such as “book” or “shop” which can operate in multiple ways. Torras contextualized how this works in a business setting.

He knows something all Industry 4.0 gurus know: the future requires more than just keyword searches.

JT: “Particularly for customer support, there are near infinite ways to state the same question. You need more than mere keywords to address this. The natural language processing platform can help you go deeper into the language and provide better answers.”

Treating Words as Symbols to Meaning

image of ancient Egyptian alphabet for article Inbenta Uses Natural Language Processing for AI-minded Chatbots inbenta chatbot
Ancient Egyptian Alphabet | Virtual Egypt

The Ancient Egyptians used a symbol and sound-based alphabet as pictured above. While Inbenta doesn’t use actual symbols, they do assign individual IDs to each word. This relates to how natural language processing increases relevancy.

JT: “Most search technology tries to operate words as a sequence of characters. Inbenta thinks of words as an internal symbolic ID. The word ‘dog’ in our system is the ID 51,202, as an example.”

Again, keywords won’t be enough to create truly lifelike, AI-based chatbots. While Inbenta uses an algorithm and natural language processing, they also use other tools to enhance their services.

You can see how the process of semantic clustering works in the following slides.

image for Inbenta Uses Natural Language Processing for AI-minded Chatbots

image for Inbenta Uses Natural Language Processing for AI-minded Chatbots

image for Inbenta Uses Natural Language Processing for AI-minded Chatbots

image for Inbenta Uses Natural Language Processing for AI-minded Chatbots inbenta chatbot
Semantic Clustering | Inbenta

The Role of Semantics in Language Processing

Inbenta uses three key techniques to differentiate themselves from other chatbots. Each technique enhances the chatbot’s ability to understand language and interpret meaning.

The first of those techniques is obviously natural language processing. But we outlined the second technique above called semantic clustering.

The process involves grouping specific types of questions under the same category. Then, based on the algorithm results, Inbenta identifies which cluster any given question most closely fits. All that’s left is to provide a relevant, useful answer.

But the final technique is something you might not know as much about: Meaning-Text Theory. This is probably the most revolutionary part of the Inbenta chatbot.

image of two brains for article Inbenta Uses Natural Language Processing for AI-minded Chatbots inbenta chatbot
singularityhub.com

How Lexical Functions Make the Best Chatbot

As described on their website, Meaning-Text Theory is “the unique theoretical linguistic framework that uses lexical functions to compute semantics”.

Aleksandr Žolkovskij and Igor Mel’čuk helped develop this AI method. It allows for a linguistic framework that augments natural language processing. The sophistication of MTT allows it to collect nearly 60 lexical functions.

That means that Inbenta’s chatbot can discern things such as synonymy or hyponymy. These fancy words mean “words that mean similar things” or “how a more generic word relates to a more specific word”.

But all of this adds up to a chatbot that helps businesses scale without ballooning costs.

image of a robot with a headset on for article Inbenta Uses Natural Language Processing for AI-minded Chatbots inbenta chatbot
PYMNTS.com

Keep Down Costs and Increase Customer Satisfaction

Torras left me with the parting wisdom that scalability matters just as much as customer interaction. As we have covered, chatbots can be a significant boon to your business. But, as you grow, you’ll need to deal with more volume and more pointed questions.

That’s where AI comes in: to relieve some of the weight of that volume.

JT: “High-quality customer support is not a ‘nice to have’. It is a HAVE to have. But if you have a customer support situation based on human labor, that would lead you to a model that is not scalable. This is why you need AI.”

With such a thorough lexical database, this AI-based natural language processing chatbot could be a basis for future androids and other life-like, inorganic creations.

For now, it makes a fantastic addition to any business with customer service needs.

What is the coolest part about the Inbenta chatbot to you?

banner ad to seo services page