An acoustic, aerodynamic and perceptual investigation of word-initial denasalization in Korean

Korean nasals /m/ and /n/ are generally considered by Korean phoneticians to be hardly different from the corresponding English sounds, but those in word-initial position are often perceived as plosives by native speakers of English. This had been noted by only a few previous observers, and investigated on a very limited scale. In this study, various experimental methods were employed in systematic analyses of the production and acoustic form of word-initial /m/ and /n/ from fluent connected speech collected from a relatively large number of informants, and corresponding perception tests were conducted with groups of Korean and English listeners. Auditory and spectrographic analyses confirmed that the segments were commonly “denasalized”. They display characteristics widely different from those of sonorant nasals, lacking the nasal formants commonly seen in spectrograms; in most cases they were more similar to voiced plosives, many tokens even showing plosivelike release bursts. Spectral analyses confirmed that denasalized nasals are significantly different from sonorant nasals throughout the whole frequency range but remain somewhat different from voiced plosives in the low and high frequency regions. Aerodynamic and accelerometer studies, which examined the consonants in CV combinations, indicated that the denasalized sounds are evidently produced with a pattern of velopharyngeal control which is different from those of sonorant nasals or of plosives. Perception tests showed overwhelmingly that the word-initial denasalized sounds are categorized as nasals by Korean listeners but as plosives by English listeners. When real voiced plosive tokens from another context are artificially moved to word-initial position, Koreans perceive these too as nasals, while English listeners’ responses are not sensitive to the context. The study shows that denasalization needs to be acknowledged as a major regular feature of spoken Korean, even though it has been largely ignored up to now. Directions for further research are outlined.