How language flows when movements don?t: An automated analysis of spontaneous discourse in Parkinson?s disease

GARCÍA, A. M., CARRILLO, F., OROZCO-ARROYAVE, J. R., TRUJILLO, N., VARGAS BONILLA, J. F., NÖTH, E., SIGMAN, M., FERNÁNDEZ SLEZAK, D., IBÁÑEZ, A. & CECCHI, G. A.

La Havana

Congreso; 18th World Congress of the International Organization of Psychophysiology; 2016

International Organization of Psychophysiology

To assess the impact of Parkinson?s disease (PD) on spontaneous discourse, we conducted computerized analyses of (the transcriptions of) brief monologues produced by 51 patients and 50 controls. The patients had been diagnosed for 11.18 years (SD = 9.16) and their motor impairments were confirmed with section III of the Movement Disorder Society-sponsored revision of the Unified Parkinson?s Disease Rating Scale (MDS-UPDRS-III). First, we explored differences in semantic fields via latent semantic analysis, a corpus-based method which generates a linear representation of the words? conceptual associations and calculates their mutual dependencies in terms of semantic distance. Second, we assessed grammatical choices using part-of-speech tagging, a technique whereby each word in a text is labeled by its grammatical function, thus revealing the groups? greater or lesser reliance on a given feature in terms of differences in tag frequency. Finally, we examined word-level repetitions with graph embedding tools, considering individual words as nodes in a network, whose links represent grammatical or semantic relationships between textual units. In addition to comparisons via t-test (with alpha values set at p < .05), we used a predictive modeling scheme (based on Support Vector Machines with Radial Basis Function kernel, with leave-one-out cross-validation) to explore the extent to which a new monologue, not previously identified as belonging to PD patients or controls, could be accurately classified as pertaining to either group. Although overall output was quantitatively similar between groups, patients relied less heavily on action-related concepts and used more subordinate structures. Also, a classification tool operating on grammatical patterns identified monologues as pertaining to patients or controls with 75% accuracy. Finally, while the incidence of dysfluent word repetitions was similar between groups, it allowed inferring the patients? level of motor impairment with 77% accuracy. Despite the absence of direct links with the patients? physiopathology, our results reveal specific discourse-level deficits associated with fronstriatal damage, beyond the possibilities of standard token-level instruments. [This work was partially supported by grants from CONICET, CONICYT/FONDECYT Regular (1130920), COLCIENCIAS (1115-545-31374 and 1115-569-33858), FONCyT-PICT 2012-0412, FONCyT-PICT 2012-1309, FONDAP 15150012, and INECO Foundation].