- Main
Recovering Meaning from "Meaningless'' Texts: Semantic Ambiguity Reduction as Recovery of Distributional Structure
Abstract
What does "blalfed'' mean in "The fuite twouch blalfed on the splelve snaitch?" Although it may seem impossible to know, people can infer word meanings from such "nonsensical'' contexts. The correct word--"sat''--was guessed correctly by 25% of our participants. We investigate people's ability to infer word meanings in such Jabberwocky texts. Participants guessed the meaning of a target word and selectively unmasked Jabberwocky words, allowing us to study the links between ambiguity, sampling choices, and accuracy of identifying the meaning of a target "nonsense'' word. Ambiguity predicted how participants sampled the text and how successfully they recover meaning, with bidirectional entropy being more predictive than entropy that only considers information preceding the target word. Accuracy increased with early information gain. These results suggest that people combine structural expectations with strategic sampling, and that semantic information is often distributed across a passage rather than localized to a single cue.