「心の仕組みとは何か」という疑問に惹かれ,答えを求めてAI分野に足を踏み入れました.出身は言語学と認知科学なので、既存のアルゴリズムを最適化することなどにはあまり興味がありません.それよりも,既存のアプローチを組み合わせたりして,AI分野が始まって以来解決されていない問題,例えば常識的知識の獲得や言語理解、倫理的な意思決定の能力などに取り組むことに興味があります.自然言語処理の研究室に勤めているため,私のほとんどの学生は標準的なNLPテーマを選択しますが,メタファーやユーモアのような,現在の機械の限界を明らかにするようなチャレンジングなテーマを薦めています.
Neurodegenerative and mental disorders significantly affect the manner of speaking, syntax, semantics and specific habits of word choice. Linguistic analysis can detect these disorders.
OBJECTIVEThe aim of this study was to examine whether speech analysis can be useful for screening test in neurology and psychiatry, due to the limited number of techniques supporting medical diagnostics in these fields. There is a need for a fast, low-cost method for analysing speech samples provided over the phone or as transcripts over the Internet.
METHODSComparing lemma frequencies in the control group recordings with lemma frequencies in speech of people diagnosed with dementia or depression allowed us to select lemmas that appear too rarely or too frequently in the speech of people affected by disorders. Moreover, the ratio of the number of lemmas to the number of words is a very good indicator of dementia.
RESULTSFor neurodegenerative and mental disorders, linguistic analysis frequently results in a more effective diagnosis than analysis of acoustic features. Linguistic changes are easily detectable in dementia, and less noticeable in depressions. By comparing features of speech samples, it was possible to create a classifier which distinguishes one group from the other. We used linguistic analysis to build a system for providing screening tests. Two methods were used to diagnose dementia. The first method is based on the observation that statements made by people with dementia have lower vocabulary variations. The percentage of lemmas (in relation to the number of words) makes it possible to detect dementia states. The second diagnostic method is based on lemma probabilities. This method was also used in depression screening tests.
CONCLUSIONSBy knowing features of speech samples recorded by the subjects and the control group, it is possible to create a classifier which distinguishes one group of recordings from the other. Linguistic changes are easily detectable in dementia, and less noticeable in depression.
CLINICALTRIALTrial Registration CT03197363; https://clinicaltrials.gov