Результаты исследований: Публикации в книгах, отчётах, сборниках, трудах конференций › статья в сборнике материалов конференции › научная › Рецензирование
Human and Machine Keyphrase Perception in Russian Text and Speech. / Гусева, Дарья Дмитриевна; Митрофанова, Ольга Александровна; Долгушин, Михаил Дмитриевич.
Speech and Computer: 26th International Conference, SPECOM 2024, Belgrade, Serbia, November 25–28, 2024, Proceedings, Part I. 2025. стр. 265-280 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) ; Том 15299 LNAI).Результаты исследований: Публикации в книгах, отчётах, сборниках, трудах конференций › статья в сборнике материалов конференции › научная › Рецензирование
}
TY - GEN
T1 - Human and Machine Keyphrase Perception in Russian Text and Speech
AU - Гусева, Дарья Дмитриевна
AU - Митрофанова, Ольга Александровна
AU - Долгушин, Михаил Дмитриевич
N1 - Conference code: 26
PY - 2025
Y1 - 2025
N2 - The article examines the perception and extraction of keyphrases in both written and spoken text. Experiments were performed on the dataset including transcripts and audio recordings of lectures by Russian-speaking participants of the project “Postnauka”. The results show that automated methods for keyphrase extraction have limited accuracy, with statistical algorithms performing the worst and generative AI models, such as ChatGPT, showing a closer resemblance to human perception. Additionally, while there is some overlap between keyphrases extracted from written and oral texts, spoken text presents greater variability. Experiments using synthesized speech indicate that listeners rely heavily on content, rather than acoustic cues, when understanding spoken text. Acoustic analysis reveals that keyphrases are distinguished by longer duration, wider pitch range, and higher energy, aligning with previous findings in other languages.
AB - The article examines the perception and extraction of keyphrases in both written and spoken text. Experiments were performed on the dataset including transcripts and audio recordings of lectures by Russian-speaking participants of the project “Postnauka”. The results show that automated methods for keyphrase extraction have limited accuracy, with statistical algorithms performing the worst and generative AI models, such as ChatGPT, showing a closer resemblance to human perception. Additionally, while there is some overlap between keyphrases extracted from written and oral texts, spoken text presents greater variability. Experiments using synthesized speech indicate that listeners rely heavily on content, rather than acoustic cues, when understanding spoken text. Acoustic analysis reveals that keyphrases are distinguished by longer duration, wider pitch range, and higher energy, aligning with previous findings in other languages.
KW - Acoustic Analysis
KW - Expert Annotation
KW - Keyphrase Extraction
KW - Perception
KW - Russian Language
UR - https://www.mendeley.com/catalogue/0036c810-4516-3284-9fe0-c1cf4319205e/
U2 - 10.1007/978-3-031-77961-9_20
DO - 10.1007/978-3-031-77961-9_20
M3 - Conference contribution
SN - 9783031779602
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 265
EP - 280
BT - Speech and Computer
T2 - 26th International Conference on Speech and Computer
Y2 - 25 November 2024 through 28 November 2024
ER -
ID: 126874264