coinstance

Misconstruction of coincidence.

A fellow instance (of a theme, concept, etc), which may be a discrete (separate) one or a coinstantiation.

Notice in particular the semantic relation which is set up between 'negative culture' and being 'totally suspicious' of change, 'looking for the rip-off', trying to 'get one over on them', 'demarcation lines', 'inflexible' and 'destructive'. We can see this as the texturing of a semantic relation of 'meronymy', i.e. a relation between the whole ('negative culture') and its parts. No dictionary would identify such a semantic relation between these expressions — the relation is textured by the manager. We can attribute this meaning-making to the manager as a social agent. And notice what the making of meaning involves here: putting existing expressions into a new relation of equivalence as co-instances of 'negative culture'. The meaning does not have a pre-existing presence in these words and expressions, it is an effect of the relations that are set up between them (Merleau-Ponty 1964).

Chevaleyre and Zucker (2001) present a general approach to upgrading learners to the multi-instance setting, apply it to RIPPER (Cohen, 1995) with good results, and briefly discuss how decision tree learners can be upgraded in a similar way. Their main point is that changing the heuristic for choosing the best split in a node is sufficient. Instead of using a heuristic based on the class distribution of instances (which is highly noisy), they propose to use the class distribution of the bags that have at least one instance in the set under consideration. In addition, in the case of decision tree induction, they state that once an instance is predicted positive by the tree, all its co-instances (instances in the same bag) should be removed from the data set. They have implemented this method in the ID3-MI system, but do not present any empirical evaluation.