[Paper Review] Analyse lexicale outill{\\'e}e de la parole transcrite de patients schizophr{\\`e}nes

Maxime Amblard, Karën Fort+3 more|arXiv (Cornell University)|Sep 2, 2015

Schizophrenia research and treatment21 references3 citations

TL;DR

This study uses NLP and lexicometric tools to analyze spoken discourse from over 375,000 words of transcribed speech in schizophrenia patients, focusing on disfluencies and lexical features (POS and lemmas). It finds that while patients produce significantly more disfluencies than controls, their lexical richness does not differ significantly, suggesting a dissociation between fluency and lexical diversity in schizophrenia-related language dysfunction.

ABSTRACT

This article details the results of analyses we conducted on the discourse of schizophrenic patients, at the oral production (disfluences) and lexical (part-of-speech and lemmas) levels. This study is part of a larger project, which includes other levels of analyses (syntax and discourse). The obtained results should help us rebut or identify new linguistic evidence participating in the manifestation of a dysfunction at these different levels. The corpus contains more than 375,000 words, its analysis therefore required that we use Natural Language Processing (NLP) and lexicometric tools. In particular, we processed disfluencies and parts-of-speech separately, which allowed us to demonstrate that if schizophrenic patients do produce more disfluencies than control, their lexical richness is not significatively different.

Motivation & Objective

• To investigate linguistic markers of schizophrenia in spoken discourse using automated tools.
• To examine disfluencies and lexical features (parts-of-speech and lemmas) in transcribed patient speech.
• To contribute to a larger project analyzing syntax and discourse structure in schizophrenia.
• To identify or refute linguistic evidence of dysfunction at multiple linguistic levels.
• To assess whether lexical richness differs significantly between schizophrenia patients and healthy controls.

Proposed method

• The study analyzes a corpus of over 375,000 words from transcribed interviews with schizophrenia patients.
• Natural Language Processing (NLP) and lexicometric tools were used for automated linguistic analysis.
• Disfluencies and lexical features (POS and lemmas) were processed separately to isolate their effects.
• The analysis included identification of hesitations, repetitions, and speech interruptions as disfluencies.
• Lexical richness was assessed via lemma frequency and part-of-speech tagging.
• Statistical comparison was made between patient speech and control speech to detect significant differences.

Experimental results

Research questions

• RQ1 Do schizophrenia patients produce more disfluencies than healthy controls in spoken discourse?
• RQ2 Is the lexical richness of schizophrenia patients significantly different from that of controls?
• RQ3 Can disfluencies and lexical features be analyzed independently to reveal distinct linguistic patterns?
• RQ4 Do linguistic markers at the lexical and fluency levels support or contradict existing models of language dysfunction in schizophrenia?
• RQ5 What role do NLP and lexicometric tools play in detecting subtle linguistic deviations in pathological discourse?

Key findings

• Schizophrenia patients produce significantly more disfluencies—such as hesitations and repetitions—than control participants.
• The rate of disfluencies is higher in patient speech, indicating potential disruptions in speech planning or execution.
• Despite higher disfluency rates, lexical richness, measured by lemma frequency, does not differ significantly between patients and controls.
• The separation of disfluency and lexical analysis revealed distinct linguistic profiles: fluency is impaired, but lexical diversity is preserved.
• The findings suggest that disfluency and lexical richness are dissociable features in the language of schizophrenia.
• The use of NLP and lexicometric tools enabled large-scale, reliable analysis of transcribed clinical speech.