corres shows which concepts travel together, and which pull apart. We do the maths. You interpret the results.
Load your matrix→corres works on departures from independence, not on counts. Two concepts that both appear everywhere will co-occur constantly, and that tells you nothing. corres looks for the pairs that appear together more often, or less often, than chance predicts. That is the difference between finding structure and rediscovering which of your codes are common.
Before corres interprets anything, it runs two tests on your corpus. The first asks whether there is enough departure from chance in your matrix to be worth analysing at all. The second resamples your documents, reruns the analysis on each half, and checks whether the same groups come back. If either test fails, corres stops and tells you which one and why — with every concept and document it set aside, itemised.
Two of the corpora we most wanted to use have been declined by our own method — one for strength, one for stability.
A method that cannot say no, cannot warn you. Knowing that before you build on it is the point.
Groups that survived the stability test. A role for every concept — which ones characterise a group, which ones bridge between groups, which ones sit apart. And the parameters and conditions each result was produced under, so the run can be reproduced.
Our own papers read groups like these as frames. That is our interpretation, and it has not been tested against an external referent — so corres returns the structure and leaves the reading to you. That part of the work is yours.
corres is free while it is in beta, and there is nothing to set up — no account, no key, no billing. Your matrix is analysed on our server by standard scientific Python libraries running a fixed procedure. No language model reads your material, and no part of the result is generated: every number comes from the same statistical code, applied the same way to every corpus. Each run reports the parameters and conditions it ran under.
corres is the first application built on the Correlating Resonance methodology. The stages upstream of it — building a typology from a corpus, and coding a full corpus against that typology — are ones we use internally to prepare our own corpora, and are in development.
corres is in beta and under active development. Some thresholds are declared provisional and are under review; every run reports the conditions it ran under.
corres is open and free during beta, and we would rather you ran your own material through it than took our word for any of this. We are most interested in the runs that fail. If corres declines your corpus, that is a result, and it is the one we most want to hear about — what you brought, what came back, and whether the diagnosis told you anything you could use.
Use the Contact button in the bottom corner of the page and tell us what happened. Name, email and what you are working on — that is all we need.
We will use your address to reply to you and to tell you when we release the upstream stages. Nothing else.