Wolfram Natural Language Understanding System™ Out Of The Box
What native features does Wolfram NLU have and how can it be utilised?
July 2021 I wrote an article on how Wolfram’s NLU and NLP capabilities can be used as a supplementary API within a chatbot. The premise was that Wolfram is especially good at data visualisation and creating images and graphs on the fly.
Fast forward…recently a video surfaced on how different technologies, including Chat-GPT, WolframAlpha, Whisper, etc can be orchestrated to form a conversational agent.
Hence I wanted to revisit Wolfram and see what NLU and NLP functionality is available in Wolfram and how it could be leveraged for a conversational agent.
For starters…the easiest way to access and experiment with Wolfram is to make use of Wolfram Cloud.
In the words of Wolfram:
Wolfram Cloud combines a state-of-the-art notebook interface with the world’s most productive programming language — scalable for programs from tiny to huge, with immediate access to a vast depth of built-in algorithms and knowledge.
Wolfram Cloud technology powers Wolfram|One, Wolfram Programming Lab,Wolfram Mathematica, Wolfram Enterprise Private Cloud and Wolfram|Alpha, as well as Wolfram instant APIs, instant Web Apps and more…
Below is an illustration of the GeoRegionValuePlot function and how data which is underpinned by geography can be displayed in a spatial fashion.
Wolfram can accommodate natural langauge queries or math input, as seen below. with NLU, a user can ask a question in natural language: “How long does it take for light to travel from the sun to the earth”. Even though the sentence is not grammatically correct below, Wolfram returns a detailed answer.
Wolfram Cloud has an extensive help panel on the right hand side, and as can be seen below, the NLP and NLU options are listed here.
Again, this is an ideal avenue for experimentation to see how natural langauge data can be processed and represented.
⭐️ Please follow me on LinkedIn for updates on Conversational AI ⭐️
⬇️ This function of Wolfram Cloud I found really interesting in the way it has similarities with LLM functionality.
Below is an example of text classification by making use of few-shot learning. In the first two examples, a few training examples are submitted for the training (yellow block), and in the green block the classification is queried.
The last to examples is of pre-existing knowledge within Wolfram which is referenced to classify flags of countries, and also an image.
The SemanticInterpretation function I find particularly interesting as it reveals the different semantic interpretations a word can have and what ambiguity can arise.
As you hover over the AmbiguityList, it is shown how the word Apple can be a company, a stock, a food type, domain, etc.
⭐️ Please follow me on LinkedIn for updates on Conversational AI ⭐️
In Conclusion
✵ Wolfram’s prowess lies with a computational intelligence and injecting that into everything at every level.
✵ It excels at answering difficult questions via natural language input, amongst others.
✵ Wolfram does not perform any dialog turn / dialog state management or conversational context.
✵ As seen above, Wolfram has NLU/NLP functionality, but this will most probably be used within the Wolfram environment as an extension to existing functionality.
✵ It is highly unlikely that Wolfram will be used as a NLU/NLP tool per se, considering the existing tools. There might be exceptions where a certain functionality (most probably some type of visualisation) is used for a specific purpose.
✵ And finally, Wolfram can act as a powerful support API within a conversational agent, to assist with specific use-cases.
⭐️ Please follow me on LinkedIn for updates on Conversational AI ⭐️
I’m currently the Chief Evangelist @ HumanFirst. I explore and write about all things at the intersection of AI and language; ranging from LLMs, Chatbots, Voicebots, Development Frameworks, Data-Centric latent spaces and more.
