Computational Modeling of an Auditory Lexical Decision Experiment Using DIANA.
Filip Nenadić, Benjamin V Tucker, Louis Ten Bosch
PMID 36000386WHAT IT FOUND
DIANA can take raw spoken English and decide whether a sound is a real word with fairly high accuracy, but its timing estimates only partly match listeners.
It also did not capture why people found some long pseudowords harder because they sounded like real words.
Key findings
01DIANA was implemented for English using raw audio from the MALD speaker, and acoustic models improved after speaker adaptation.
02At the selected threshold, DIANA classified real words at 87.92% and pseudowords at 76.44%, and it was more accurate for longer pseudowords than short ones.
03DIANA's response-time estimates matched participant response times only moderately, and the best fit required subtracting time after word offset, which conflicts with the model's own assumptions.
STILL TO COME
How it was doneWhat they foundWhat it means for SLPs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The study tests a computational model, not patients, so it cannot show that a clinical procedure works. The model was built for a single English speaker and a laboratory dataset, so it may not reflect varied speakers, accents, noise, or clinical populations. The model's acoustic input and competitor selection required simplifications, and the model's timing mechanism did not fully match listener behavior. The model ignores word meaning and morphology, which affects how people judge pseudowords. The paper is about model development, so the findings are technical and not directly usable at the bedside.
Declared interests
The research was funded by the Social Sciences and Humanities Research Council of Canada.
The easy way to misread this
Do not treat DIANA as a clinical tool for assessing spoken word recognition. This paper describes a model-building study on laboratory recordings, and its timing estimates did not fully match listener behavior.