Term-Chunk correlations tell us how strongly terms are connected with each of the chunks. Taking a single term, its term-chunk correlations provide a measure of where in the corpus its use is significant.
Terms will frequently have strong term-chunk correlations with locations where the term is not present. Words belong to semantic networks and LSA Term-chunk correlations actually detect the presence of the word's semantic network.
When a term is present in a chunk and has a strong correlation, that location will normally be very informative about the usage of that term in the corpus. Locations with weaker correlations will provide less information about its usage.
The user can construct a query based on terms that works in a similar way. See Query with Terms