Assessing problems retrieving words and their meanings with machine learning

What is it about?

After left hemisphere stroke, up to half of people experience language problems, including problems producing words. Sometimes people after stroke produce words that are close in meaning but not what they intended to say, for example saying “violin” when they meant to say harp. This suggests a problem retrieving information related to words and their meanings. However, evaluating these kinds of naming mistakes is typically subjective, takes time, and does not capture how close the speaker’s naming attempt was to what they intended. To address these challenges, we tested object naming in a group of 105 people within a few days of a left-hemisphere stroke. We showed that a fast and objective computational approach, word2vec, provided an excellent estimate of the problems a person has retrieving information related to words and meanings while also outperforming human judgements. This improved approach to assessing naming will help us better understand how naming errors happen and develop approaches to improve language problems after stroke.

Featured Image

Photo by David Pupaza on Unsplash

Photo by David Pupaza on Unsplash

Why is it important?

Our validated word2vec-based method provides both theorists and clinicians an easy to implement, accurate and consistent tool to measure problems retrieving information related to words and their meanings during naming.