When a panel carries out a brainstorming session on an issue or research question, it is often difficult to apply text summarization to the brainstorming notes. This either gives the power of editing conclusions to the panel leader, which may favor some ideas, and/or slows down considerably the brainstorming process so that insufficient refinements can be carried out. It is in this context that natural language analysis, relying on linguistic resources, can focus on semantic classes and causal relations, to generate from the set of brainstorming notes a smaller set of hypothesis rules applying to the field being studied. In some cases, scenario analysis can even be quantified. The present paper documents in minute detail, in a specific case, how brainstorming note analysis and rule condensation can be made. The brainstorming raw material are reproduced, as well as all intermediate results up to and including the validation rules. The case is taken from the European “Megaprojects” COST collaboration which deals with the effective design of very large projects, esp. in infrastructure, energy, etc., called Megaprojects. The panel the results of which were studied was held in Bratislava in May 2012, and analyzed in Milan in July 2012. The condensed hypothesis rules produced by natural language analysis were further validated in and by six real “megaprojects,” to yield ultimately a set of general rules, open for later refinement by the same process. Taking the syntactically edited brainstorming notes from the Bratislava panel (Section 1), triggered by an unfocussed question on Megaprojects in general, and further manual editing/preparation/domain adaptation (Section 2) as input, the automated proprietary tool generates several results. They include in Section 3 and Section 4 the nominal group clustering tree with the assumed root structure. Section 5 gives the aggregated hypothesis rules organized by aggregation mechanism. The results, which are the hypothesis rules ranked by relevant note counts are given in Section 6. The note counts linked to specific hypothesis rules are given in Section 7 for quality control to avoid dominance of any specific brainstorming notes. The hypothesis rules are subject by two ratings (Section 8) from 6 independent Energy WG cases, by case authors and/or project management. In summary, the natural language analysis of 70 sentences produced by the unfocussed brainstorming panel, and subsequent rating of the derived hypothesis from 6 independent cases, produces the following general rules about Megaprojects: Hypothesis rules validated with high note counts: A: Good management and good outcome are correlated; C: Links and incentives must be created for success for the different categories of Users and Builders; Hypothesis rules validated with average or low note counts; F: Policy AND Regulations are linked to execution AND outcome; G: The deployment context must be aligned with local benefits; H: Performance is improved by management; I: The same actors are found across Europe; J: Megaproject characteristics are linked to outcome. Three hypothesis rules are not validated. Subsequent refinement of these rules, or the generation of new rules, for example on priorities regarding the execution, or success, or other characteristics of Megaprojects, would require a more focused second brainstorming with one initial proposition/statement to which panelists would have to react in view of producing focused comments (to be analyzed in the same way). This is standard in sequential brainstorming analysis with refinements. Compared to state-of-the-art research in text summarization, the problem and solution presented here offer the advantage of better flexibility in view of domain specific terminology or semantic meanings
