If you have listened to Korean and thought there’s something odd about the pronunciation of the word-initial nasals /n m/, you’re not alone. In fact, you probably know more than the average native speaker about the peculiarity of Korean nasals! I bet that an absolute majority of Koreans are totally unaware of it, something which is so striking to many Korean learners.
So, yes, you were spot on when you thought you heard something like [dugu] or even [t(h)ugu] instead of /nugu/ ‘who’ (누구), and [bwʌrago], [p(h)wʌrago] instead of /mwʌrago/ ‘what?’ (뭐라고), while watching your favourite K-drama. In other words, nasals in Koreans lose nasality and sometimes even voicing in word-initial position. It’s only very recently that word-initial denasalisation became recognised by Korean phoneticians as a regular feature of spoken Korean and became a topic of systematic investigation (Yoshida 2008; Y-S Kim 2011).
Korean Denasalisation shows patterns associated with so-called domain-initial strengthening (Cho & Keating 2001). Domain-initial strengthening refers to the phenomenon in which domain-initial consonants are pronounced with longer, greater articulatory gestures. The domain here refers to any prosodic domain, which can be defined as the unit of speech larger than segments (vowels and consonants) such as syllable, phonological word (PW), accentual phrase (AP), intonational phrase (IP), and utterance. Domain-initial strengthening is found with different types of sounds in different languages including French, English, Estonian, Tamil, Taiwanese, Dutch, and Korean. For example, English /t/ and /h/ show a longer aspiration (the ‘h’ sound in e.g. ‘thick’, which creates a puff of air; this isn’t found in e.g. ‘stick’) and a larger glottal gesture in phrase-initial position than in phrase-medial position (Pierrehumbert & Talkin 1992).
Now, here’s why domain-initial strengthening is so cool. The effects of domain-initial strengthening are cumulative, so that the larger the prosodic domain of the initial segment, the greater the effects. Using domain-initial denasalisation as an example, IP (or large phrase, in plain English) -initial nasals are more denasalised than AP (a unit made up of one or two phonological words in Korean) –initial nasals, and AP-initial nasals are in turn more denasalised than PW-initial nasals, and so on. Let’s look at some examples:
1(a) /ob*a IP[mandu mʌgɯrʌga]/
‘Obba (what you call an older brother), are you going to eat dumplings?’
오빠, 만두 먹으러 가?
1(b) /ʌmma ob*a AP[mandu mʌgɯrʌga]/
‘Mom, is Obba going to eat dumplings?
엄마, 오빠 만두 먹으러 가?
1(c) /ʌmma ab*aga AP[ob*a pw[mandu]] mʌgɯrʌga]/
‘Mom, is Dad going to eat “Obba Dumpling” (product name)?
엄마, 아빠가 오빠 만두 먹으러 가?
All these sentences contain the word /mandu/ ‘dumpling’, with the word-initial /m/. The key difference is that the highest possible position of /m/ is the IP-initial position (though, of course, it is also syllable-initial, word-initial, and so on) in (a), AP-initial position in (b), and PW-initial position in (c). Because the effect of Korean denasalisation is cumulative, the extent of denasalisation is (a)>(b)>(c). I would be surprised if an English listener heard the /m/ in (c) as anything other than [m]. But the /m/ in (a) is likely to be completely denasalised and devoiced as [p], and may even be slightly aspirated [ph]!
As what’s described so far is a relatively new discovery, there is a lot to be known about domain-initial denasalisation in Korean and domain-initial strengthening in general. One of the most exciting questions to ask next is what role domain-initial strengthening plays in perception. Because we now know that the pronunciation of consonants is not invariant but shows fine-grained variation as a function of prosodic position, we can hypothesise that this information can be used by listeners as a cue to a prosodic boundary. Consider the following examples:
2(a) IP[While eating], IP[my dog, my cat and I watched the TV].
2(b) IP[While eating my dog], IP[my cat and I watched the TV].
I’m using English examples for a quick illustration, but similar examples can be found easily in Korean. For now, let’s assume that English has domain-initial denasalisation. We would then expect the /m/ in (a) to be denasalised to a greater extent than in (b), because it is in IP-initial position. Given the pattern, are listeners able to use denasalisation as a cue in speech perception to disambiguate these sentences, in addition to pitch and duration cues? This is exactly what I’m trying to find out during my fieldwork in Korea in the next two months, so we’ll see!
 Slant brackets are for phonemes, the abstract mental representation of a segment in the speaker’s mind. If you like, this is what the speaker thinks they are saying. On the other hand, square brackets are for (allo)phones, the actual realisation of phonemes, which may be slightly different every time they are uttered and from what the speaker thinks they are pronouncing.
Cho, T., & Keating, P. A. (2001). Articulatory and acoustic studies on domain-initial strengthening in Korean. Journal of Phonetics, 29(2), 155–190.
Kim, Y. S. (2011). An acoustic, aerodynamic and perceptual investigation of word-initial denasalization in Korean. University College London.
Pierrehumbert, J., & Talkin, D. (1992). Lenition of /h/ and glottal stop. Papers in Laboratory Phonology II: Gesture, Segment, Prosody, 90–117.
Yoshida, K. (2008). Phonetic implementation of Korean denasalization and its variation related to prosody. IULC Working Papers, 8(1).