Premium
Automated semantic speech analysis in AD and lvPPA
Author(s) -
Nevler Naomi,
Ash Sharon,
Cho Sunghye,
Shellikeri Sanjana,
Parjane Natalia,
Irwin David J,
Liberman Mark Y,
Grossman Murray
Publication year - 2020
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1002/alz.045300
Subject(s) - morpheme , primary progressive aphasia , audiology , psychology , cognition , connected speech , concreteness , computer science , medicine , cognitive psychology , natural language processing , disease , pathology , dementia , neuroscience , frontotemporal dementia
Background Speech is a complex behavior where restricted cognitive impairment in one domain may affect patients’ performance, thus providing opportunities to identify disease markers. A limited number of studies have examined natural connected speech in typical and atypical Alzheimer’s disease (AD). Here we implement automated language processing to characterize some semantic properties of speech in speakers with typical AD and its linguistic variant, logopenic variant primary progressive aphasia (lvPPA). Method We analyzed transcripts of natural speech elicited by a picture description task from AD, lvPPA and matched control speakers (Table1). We automatically parsed and tagged each word and then automatically scored for word length (in phonemes and morphemes), frequency, familiarity, semantic diversity and concreteness based on published norms. We compared these features across groups, covarying for age, sex, education and Mini Mental Status Examination (MMSE) score. We correlated each acoustic measure with MMSE scores collected within 6 months of the speech samples. Result AD and lvPPA speakers differed from controls in all semantic scores except for word familiarity and semantic diversity: they used higher frequency words (p<0.001) that were also shorter (phoneme count, p<0.001; morpheme count, p<0.001), more ambiguous (p=0.028) and abstract (p=0.005) and acquired at a younger age compared with control speakers. lvPPA speakers did not differ from AD speakers in their words’ length in terms of the number of phonemes. However, the number of morphemes was less in lvPPA (mean 1.104±0.037) compared with AD (mean 1.129±0.038) , p=0.002). They also showed a trend towards using higher frequency words (p=0.057), acquired at a younger age (p=0.05) compared with AD speakers. lvPPAs differed from control speakers in their semantic diversity (p=0.02). All scores reflecting impaired performance correlated with MMSE scores in patient speakers. Conclusion In natural speech AD and lvPPA speakers select words that are shorter, more frequent, semantically ambiguous and less concrete. lvPPA speakers show a specific predilection to words with fewer morphemes and higher frequency, acquired at an earlier age. These deficits correlate with the degree of cognitive impairment and may negatively impact communicational efficacy in these patients as they deteriorate.