Why machine learning can't understand human language

The current trend in the artificial intelligence community is poised on solving problems mostly by creating bigger neural networks. This approach has led to incremental improvements in many fields, including natural language processing.

Today, we have deep learning models that can generate article-length sequences of text, answer science exam questions, write software source code, and answer basic customer service queries.

But the deeper problem of natural language understanding (NLU)—dealing with the iceberg hiding under words and sentences—remains largely unsolved. And this is because the AI community has abandoned knowledge-based systems, argue Marjorie McShane and Sergei Nirenburg in their new book Linguistics for the Age of AI.

Linguistics for the Age of AI provides an in-depth exploration of the shortcomings of current NLU systems and provides a roadmap for creating intelligent agents that can deal with language and all its intricacies in a way that resembles humans.

McShane discussed the book and NLU with TechTalks. Read the article here.

For more on AI and natural language:

• AI’s struggle to reach “understanding” and “meaning”

• Practical NLP: The perfect guide for executives and machine learning practitioners

• The sweet spot of deep learning and natural language processing