Does one chunk correspond to the raw signal of a single read read? Why does messing with this parameter affect basecalling results?
If you have access to the nanopore community website you can read more about this here: https://community.nanoporetech.com/posts/not-chunky-enough (I'm not sure to which extent I can copy paste from there so I won't).
A TLDR version: basecalling doesn't happen simultaneously for an entire read, but it is broken in segments which are processed separately and combined afterward. Those segments are chunks. Basecalling results are affected mostly by low-complexity regions, in which a heavily biased region (in terms of nucleotide content) leads to artifacts in the per-chunk basecalling. Those issues are solved by taking a chunking-size larger than most low-complexity regions.