Building natural language by way of crossword puzzles could be an approach used for robots or even web intelligence. The approach is an inspiration taken from autism especially from the movie Mercury Rising. Being able to produce a level of intelligence where crossword puzzles could be utilized as way of open question/answering. In process, devising a way for reinforcement learning by which a softbot or even a robot can learn, comprehend, and understand the English language. This could even be taken further to understand the every day usage of slang words and even sentiments within text and speech. Furthermore, extending the approach to multiple languages. Word puzzles in general are a good logical way of building iterative inference based learning which allows for a continuous feedback loop. Even applying a bio-inspired intelligence could be an option here. However, this may be still slow in performance for the real-world as evolutionary computation does seem to have a very slow learning rate in general even on the margins of simple representations. The intelligence could even be encapsulated with memory for grammar, visual cues, speech metaphors, vocabulary, spelling, word disambiguation, and a lot more. And, a variety of context and complexities would be a way of adding further challenges. Going further, the approach can be utilized to build more context specific computational approaches for game shows as well as to improve variable statistical inference.
