LEE, Seunghun (ICU) & KAWAHARA, Shigeto (Keio)
Korean vocative truncation and Information Theory: A perspective from Message-based phonology.
INTRODUCTION: Recent studies have explored the usefulness of Shannon’s (1948) Information Theory in explaining linguistic patterns, ranging from phonetic implementation patterns (Aylett & Turk 1994) to syntactic patterns (Jaeger 2010). Aylett & Turk (2004) show that in English, more predictable vowels are shorter and more centralized. Jaeger (2010) demonstrates that predictability may affect syntactic patterns in that speakers attempt to distribute information more or less consistently across the signal.
Within the domain of the morphology-phonology interface, Shaw et al. (2014) show that in the Modern Standard Chinese truncation compounding pattern, what survives in truncation tends to be those segments that are informative; i.e. those segments that allow listeners to recover what the original, untruncated words are. Shaw et al. (2014) moreover show that a purely grammatical explanation, such as the one based on morphosyntactic headhood, does not explain the truncation pattern very well. These studies have been crystalized into the theory of “Message-based Phonology” (Hall, Hume, Jaeger & Wedel 2016), in which P(message| signal, context) (termed “probability of accurate message transmission”) plays a crucial role. This conditional probability is the probability of the listener retrieving the correct message given its signal and the context. Message-based Phonology argues that effective message transfer is achieved when P(message| signal, context) is kept consistently high. This paper analyzes the Korean vocative truncation pattern from this perspective, and shows that what survives in truncation is those that would keep high P(message| signal, context).
DATA: Korean first names consist of two part: one part is shared by the same generation of siblings and cousins, and the other part is unique to each person. The order between these two elements alternates between one generation to the other. Examples in (1) illustrate, where the underlined portions are those that represent the same generation.
(1) First generation Second generation Third generation …. hong+hui [hoŋ.hɨi] jae+eun [t͡ʃe.ɨn] min+su [min.su]
dong+hui [doŋ.hɨi] jae+young [t͡ʃe.jɤŋ] in+su [in.su] seok+hui [sɤk.hɨi] jae+hun [t͡ʃe.hun] hui+su [hi.su] yang+hui [jaŋ.hɨi] jae+jun [t͡ʃe.dʒun]
ja+hui [t͡ʃa.hɨi] jae+u [t͡ʃe.u]
What happens in vocative truncation is to take the unique part of the names, and add [(j)a] to them, as in (2). [j] appears when the unique part ends with a vowel.
(2) hong-a [hoŋ.a] eun-a [ɨ.na] min-a [mi.na] dong-a [doŋ.a] young-a [jɤŋ.a] in-a [i.na] seok-a [sɤ.ga] hun-a [hu.na] hui-ja [hi.ja] ja-ya [t͡ʃa.ja] u-ya [u.ja]
So why do unique parts survive in truncation? In other morphological truncation patterns in Korean and Japanese, they usually retain first syllables, as in (3):
(3) Korean Japanese
[haŋkuk] + [ɨnheŋ] -> [hanɨn] [paasonaru]+[kompuutaa] => [pasokon]
This preference toward initial materials is easy to model, as an effect of positional faithfulness constraints to initial portions (Beckman 1998).
LEE, Seunghun (ICU) & KAWAHARA, Shigeto (Keio)
How do we then explain the patterns in (2)? We maintain that the same principle that Shaw et al. (2014) argue for—keep P(message| signal, context) high (Hall et al. 2016)—is at work. Take the leftmost column in (1) as an example. Given the context of the [hui] family, the probability of [hong] picking out the right denotation (/hong+hui/) is 1; on the other hand, the probability of [hui] picking out the right denotation is 1/5 (or to generalize, 1/n where n is the number of family members). In terms of entropy, [hui] has 0 bit, whereas [hong] has 2.3 bits (-log2(0.2)). In truncation, it is crucial that the listener can recover the original form before truncation. Thus eliminating the element with no entropy—the element that is predictable— does not lower P(message| signal, context); eliminating the portion with high entropy would result in lowering P(message| signal, context).
Another interesting aspect of this truncation is the fact that when you meet a new person, and hence do not know which part of the name is the unique portion, truncation is impossible, or at least highly unlikely. In other words, this morphological process is blocked when there is a danger of lowering P(message| signal, context). This blockage pattern is a reminiscent of M-PARSE effect (Prince and Smolensky 1993 et seq; see the contributions in Blaho and Rice 2010), where phonological and other restrictions can block a morphological process.
One could argue that what survives in truncation is morphological head, and hence there is no place for the information value of the element to play a role. However, saying that what survives in truncation is a head is restating the observation, and is not explanatory.
There is independent piece of evidence that Korean speakers keep P(message| signal, context) high when it comes to morphological truncation. Consider (4):
(4) [i-de] (from [ihwajɤ-de]), [jɤn-de] (from [yɤnse-de]), [ko-de] (from [korjɤ-de])
Some university names can be truncated, by taking the first syllable of the university name and keeping [de], which means ‘university’. However, ‘Seoul National University’ and
‘Seoul Yeodae (Seoul Woman’s University)’ cannot be truncated, to either [sɤ+de] or [ul+de], because these truncated forms may represent other universities; i.e. [sɤwon+de] and [ulsan+de].
We also find some admittedly sporadic examples from Japanese that show that informativity plays a role in morphological truncation. In Japanese, the preference toward keeping initial elements is very strong. However, there are a few examples that violate this preference: [suta+kon] from [misutaa kontesuto] ‘Male beauty pageant’, cf. [misu+kon] from [misu kotensuto]. Keeping the initial syllables of [misu] from [misutaa] would result in ambiguity with [misu], thereby lowering P(message| signal, context). Another example is [nai+kara] from [agenai karaage] ‘non-deep-fried fried chicken’. From the informativity perspective, what is important about this product is the fact that it is not deep-fried (despite tasting like fried chicken), and hence [nai] ‘not’ survives.
CONCLUSION: Some morphemes carry more information content than others. Korean speakers are aware of this difference, and in making new words based on truncation, they keep informative morphemes intact. In Message-based Phonology, informativity in this sense can be formally expressed as P(message| signal, context). The vocative truncation is blocked when it runs the risk of resulting in low P(message| signal, context). Korean vocative truncation thus provide evidence that speakers keep P(message| signal, context) high. Many examples discussed in Hall et al. (2016) are “phonetic-y” phenomena, but we maintain that the same principle is relevant in morphophonological patterns as well, a la Shaw et al (2014).
