In everyday communication human beings rely on linguistic regularities to answer the semasiological question “what does the word X mean?”. Yet, we also need the opposite, that is answering the onomasiological question “how do I express the concept X?”. In this study, we explored whether widespread systematicity between form and meaning estimated through Linear Discriminative Learning can be leveraged to generate novel words from predefined semantic information.

We conducted three experiments to explore whether human participants were able to identify a generated pseudoword resulting from one single word prompt (i.e., a “pseudo-synonym”, Experiment 1), a pair of word prompts whose combined meaning produced a certain pseudoword (Experiment 2), and a generated pseudoword resulting from the combined meaning of a pair of word prompts (Experiment 3).

Participants proved to be able to identify the correct pseudoword(s)-prompt(s) association, and semantic similarity between response options proved to explain human performance over and above surface similarity.

These results suggest that regularities between form and meaning not only support language comprehension but also play a role in the formulation of novel ideas in speech production. Mappings from a semantic space to a form space provide a new tool for simulating the production of (and actually generating) novel terms.