The article spells out in some detail why we have found it so difficult to build algorithms capable of understanding spoken and written language to the extent of being able to engage in a proper conversation with humans. By way of benchmarking, the article cites Siri and Watson as able to engage in relatively simple verbal interaction, but no more. In other words, today’s natural language processing is not progressing to an adequate level of sophistication despite the great strides made in recent years. Still, as the application of machine learning techniques becomes more sophisticated in terms of both depth of technique and also their combination, progress in this area will continue to be made. In particular, enabling such language systems to interact more with the real world will hugely improve their ability to infer meaning where otherwise ambiguity would prevail.
Link to article: https://www.technologyreview.com/s/602094/ais-language-problem/