Divergent thinking (DT) is a fundamental part of creative ideation. Understanding its role in cognition and its attainment through language technology can provide the scaffolding to enhance creative endeavors. This study is a proof of concept on the automatic generation of keyword responses as found on the AUT (Alternative Uses Task), a test commonly used to measure DT. Within a psychometric perspective, we propose a frequency-based simple word co-occurrence method called Co-OBM (co-occurrence-based method). Drawing from Natural Language Processing (NLP) we choose the adequate parameter settings for this task, including part-of-speech tagging (POS), word form, association measure, minimum occurrence in corpus and window size. Through our experiments, we show how most of the popular responses to AUT can be identified in a short word co-occurrence list, together with some of the least frequent, and usually more creative, responses, how this outcome is not random but based on linguistic patterns (Experiment 1); and how Co-OBM output can enhance the performance of subjects' AUT responses (Experiment 2). This is relevant since word co-occurrence is at the core of current language models. Our work aims to reflect and provide technical and empirical foundations for the development of distributional language models for creative purposes.
