Last week, Fonolo hosted an insightful live discussion on the current state of AI and Natural Language Processing (NLP). The expert panel discussed how NLP relates to the general field of AI, the pros and cons of different approaches to NLP, and more. To save you some time, we’ve extracted video snippets showcasing highlights from the discussion.
Under AI’s Umbrella
NLP, or natural language processing, is a fairly specific concept. It is the process of extracting intent from human language. On the other hand, the term “AI” is an umbrella that covers an ill-defined group of technologies. Some approaches to NLP use machine learning, and so are more qualified to fit under that umbrella, but some are rules-based and don’t fit as well. The connection with AI also sets sometimes unrealistic expectations for what NLP can do.
Should we continue to call NLP a subset of AI? How about “NLP” vs “NLU”? What are the right ways to talk about this technology?
Machine Learning vs. Linguistic Rules
The advantage of NLP is that it allows open-ended dialogue with a customer. In theory, this leads to higher completion rates with self-serve processes. Ultimately, we’re trying to serve a customer without requiring a human in the loop. There are two basic approaches to get there, one based on machine learning and one based on linguistic rules.
What are the strengths and weaknesses of each? Which of those two approaches is winning in the market right now?
The Advantages of “Speech to Intent”
For the voice use-case, another option with NLP is to bypass entirely the transcription stage and instead go straight from the audio signal to intent. This is a newer approach that is leading to some very impressive results.
What are the advantages and disadvantages of going straight from speech to intent?