sunflower: an R package for handling multiple response attempts and conducting error analysis in aphasia and related disorders

Abstract

Manual classification of production errors and the allocation of speech/spelling scores are time-consuming, laborious and error-prone tasks, even when conducted by clinicians and specialized researchers. Here we present sunflower, an R package developed to improve the analysis of language production quality for Spanish data. The package offers various functions, including (1) managing dataframes containing single responses and multiple-attempt responses, (2) conducting formal similarity analyses on words as well as positional accuracy data analyses within words, and (3) the classification of errors by considering lexicality, formal similarity and semantic similarity indexes, which are obtained by means of different algorithms and artificial intelligence techniques such as word2vec. The applications of sunflower, which is the first open-source package of its kind, include assessing whether production quality improves over the course of multiple attempts, and identifying which aspects of an individual’s productions are most impacted by their impairments. Other potential applications include the analysis of whether improvements arise in a patient’s production quality after a given treatment, distinguishing between cases of apraxia of speech and conduction aphasia, as well as simply using the package to improve and speed up the classification of speech/spelling errors with large datasets through automation.