Semantic consistency of the sub-graph
To perform the evaluation, each of the 6,342 stimulus-response pairs used to build the network were manually evaluated. The evaluation was necessary because of the observation that the free word association experiment may produce so-called clang associations, i.e. words that sound like a stimulus or rhyme with the stimulus, e.g. house - mouse and the idiom completion associations, e.g. white - house, which forms a multipart lexical unit and, therefore, does not reflect the meaning relation between stimulus and response [CLA 70]. We had expanded this observation treating all associations which introduce proper names, e.g. river - Thames and the not so frequent deictic association, e.g. girl - me as non-semantic.
The evaluation is as follows. If the stimulus is semantically related to the response as in dom - sciana (house - wall), the pair is marked as semantic, otherwise the pair is considered to be a non-semantic one, e.g. gora - Tatry (mountain - the proper name) or dom - moj (house - my).
Then, the sub-graph nodes were evaluated consecutively along a path in the following way. If the two connected nodes matched a stimulus-response pair marked as semantic, then the right node was marked as semantic - Sn. If the two connected nodes matched a non-semantic stimulus-response pair, then the right node was marked as non-semantic one - nSn. If the two connected nodes did not match any stimulus-response pair, then both nodes were marked as nSn, except the stimulus node which is in principle a semantic one. After the final pair of a path was evaluated, the evaluation of the connection of the start node (stimulus) and the end node of the path was evaluated to check the semantic consistency of the path. As a result, a non-semantic node nSn is considered to be any end node (association) that does not have a semantic relation to the start node (stimulus), even if it has a semantic relation with a preceding node, e.g. the path: krzeslo - stol - szwedzki (chair - table - Swedish), where pairs krzeslo - stol and stol - szwedzki enter into a semantic relation, but the stimulus krzeslo “chair” does not enter into a semantic relationship with the association szwedzki “Swedish”.
Matching the sub-graph and a text collection
To assess how the extracted sub-graph relates to a text collection, we have to match each text that contains a particular stimulus against the subgraph extracted for this stimulus. Then, we have to count the network nodes (words) that were recognized both in the texts and in the sub-graph SnT. After that, we have to match the entire set of direct associations with a particular stimulus which is present in the network against the texts. This is performed in order to recognize network nodes (words) which are present in the network but were rejected by the algorithm, and therefore, are not present in the sub-graph TnS.