Spoken Recall Reveals Lexical and Mnemonic Function Differences in Temporal Lobe Epilepsy Patients

Abstract

Background and Objectives Word-finding difficulty is common in healthy aging and in neurologic disorders, including temporal lobe epilepsy (TLE) and early Alzheimer disease. Standard language measures have limited sensitivity for detecting subtle or longitudinal changes in spontaneous speech. We examined whether natural language processing and acoustic analysis of spoken biographical recall could identify lexical and temporal speech features associated with language and memory performance in TLE. Methods We conducted a cross-sectional observational study of spoken recall during a Famous Faces biographical memory task. Adults with TLE and healthy controls (HCs) viewed 20 famous faces and spontaneously recalled biographical details. Speech was transcribed and diarized using automated tools. Lexical measures included word counts and lexical index (ratio of rare to common words). Acoustic measures included utterance and pause duration and pause frequency. Features were compared between groups and correlated with neuropsychological measures, including Montreal Cognitive Assessment (MoCA), Boston Naming Test (BNT), delayed recall, education, and biographical recall accuracy. Results Eighty-one adults participated (51 TLE, 30 HCs). Lexical measures did not differ between groups. In TLE, lexical index correlated with BNT performance (r□=0.62) and MoCA score (r□=0.35). Compared with HCs, participants with TLE produced shorter utterances (5.54 ±2.50 vs. 6.59 ±2.63; Cohen’s d=0.41, 95% CI -0.05 to 0.86, p=0.041), shorter pauses (0.61 ±0.21 vs 0.67 ±0.22, Cohen’s d=0.29, 95% CI -0.16 to 0.74, p=0.043), and more frequent pauses (10.81 ±3.30 vs 9.29 ±3.59, Cohen’s d=-0.45, 95% CI -0.90 to 0.01, p=0.036). Higher education was associated with longer utterances, longer pauses, and lower pause frequency, without evidence of a diagnosis-by-education interaction. Faster utterance rate was associated with better biographical recall in both groups, while higher pause rate was associated with worse recall in TLE. Discussion Speech-derived lexical and temporal features from naturalistic recall capture clinically relevant variation in language and memory-related performance. Although lexical output did not distinguish TLE from HCs, lexical richness tracked naming and global cognition in TLE, while temporal speech features related to recall performance. These findings support the potential of automated speech analysis as a digital behavioral biomarker for word-finding difficulty in neurologic populations.