I am experimenting with statistical next word prediction. My goal is to build a system that can accurately answer a question given a corpus of text. For now, my system is producing interesting summaries of content. Below is a summary of my post on the joy in listening to songs you used to love from a few days ago, generated by the tool I am building.
An aside is that this output is statistically modelled, which means that it could be replicated every time you ran the program. It makes me wonder how far we can go with statistics-based summarization like this. Below, you will see the fidelity of the content is low, verging on incomprehensibility in places. Nevertheless the output is intriguing!
needed did, brought Maybe even which get urge listen got suddenly coffee make preparing cup preparing was the accompanying Cleopatra Ballad rhythm, repeat.
music, The obliged, Lumineers.
opening Spotify play track.
Gloria one songs another Stone started playing, Julia Stone.