Assessing Hypotheses in Multi-Agent Systems for Natural Language Processing