Vous êtes sur la page 1sur 17

Available online at www.sciencedirect.

com

Journal of Pragmatics 42 (2010) 10001016 www.elsevier.com/locate/pragma

The historical development of Korean siph- to think into markers of desire, inference, and similarity
Minju Kim *
Department of Modern Languages, Claremont McKenna College, 850 Columbia Avenue, Claremont, CA 91711-6420, United States Received 26 January 2007; received in revised form 14 August 2009; accepted 22 August 2009

Abstract Using diachronic corpus data, I investigate the historical development of the Korean psychological predicate siph- to think to oneself, which engendered the desiderative -ko siph- want to and -myen siphi- would be great if. . . ; the inferential -ka siph-, -tus siph-, and -seng siph-, which all mean probably; and the connective siphi similar to or as. Before the 15th C, the desiderative ko siph- and the inferential -ka siph- were rst grammaticalized within their usual contexts of use; additional desiderative and inferential constructions with siph- were subsequently created through replacement of one morpheme of these original constructions, modeled after existing constructions with similar functions. Cross-linguistically, markers of similarity often develop into markers of probability. However, the Korean connective siphi similar to developed from the probability meaning of siph-. Although it seems to defy the unidirectionality hypothesis that grammaticalization moves from more concrete to more abstract meaning (Traugott, 2002), I demonstrate that the development of siphi was inuenced by analogy to the existing constructions kathi and tusi similar to. Based on Kiparsky (forthcoming) and Lehmann (2004), I propose that some cases of degrammaticalization that were inuenced by analogy specic to given languages may not be legitimate counterexamples to the unidirectionality hypothesis. # 2009 Elsevier B.V. All rights reserved.
Keywords: Korean; Desideratives; Inferentiality; Unidirectionality hypothesis; Semantic extension; Psychological predicate; (De)grammaticalization; Analogy

1. Introduction Grammaticalization is typically dened as a change through which a lexical item in certain uses becomes a grammatical item, or through which a grammatical item becomes more grammatical (Hopper and Traugott, 1993:2).1 In the classical view of grammaticalization, peoples pragmatic understanding of linguistic structures in the communicative use of language is incorporated in the shaping of their grammar and language. More specically, generalized pragmatic inferences from contexts of use often transform into new grammatical functions (Traugott and
* Tel.: +1 909 607 2503; fax: +1 909 621 8419. E-mail address: minju.kim@cmc.edu. 1 As will be discussed in section 7, Hopper and Traugott changed the denition of grammaticalization in their revised book to mean a change whereby lexical items and constructions come in certain linguistic contexts to serve grammatical functions and, once grammaticalized, continue to develop new grammatical functions (Hopper and Traugott, 2003:xv). Nevertheless, I have cited the original denition from Hopper and Traugott (1993) because when examples of degrammaticalization are discussed, most of the cases under question are raised on the basis of the original denition. 0378-2166/$ see front matter # 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.pragma.2009.08.010

M. Kim / Journal of Pragmatics 42 (2010) 10001016

1001

Dasher, 2002; Hopper and Traugott, 2003 [1993]). For this reason, the elds of pragmatics and grammaticalization share important similarities, including their preference for naturally occurring data and their emphasis on the communicative use of language. With close attention to semantic-pragmatic aspects, this article will use the theoretical framework of grammaticalization in conjunction with diachronic corpus data to investigate the historical development of the Korean predicate siph- to think to oneself. Currently, siph- is used in three grammatical constructions: as in (1) and (2), in constructions with an auxiliary predicate siph- marking the desiderative meaning want to (-ko siph-) and the inferential meaning probably (-ka siph-); and as in (3), in the connective (or conjunctive sufx) siphi as. The connective siphi is derived from the combination of the auxiliary predicate siph- and the adverbial sufx i, and is used with the declarative ender -ta (e.g., V-ta siphi). (1) Desiderative want to na-nun icey ca-ko siph-ta I-TOP now sleep-ko siph-DEC2 I want to sleep now. Inferential probably Ol-hay hyung-cak-un kamwum ttaymwun-i-n-ka siph-ta this-year bad harvest-TOP drought reason-COP-RL:PRES-ka siph-DEC This years bad harvest has probably been due to the drought. Connective as ney-ka po-ass-ta siphi ku-nun imi ttena-(a)ss-ta. you-NOM see-PAST-ta siphi he-TOP already leave-PAST-DEC As you saw, he had already left.

(2)

(3)

Examples of the desiderative construction -ko siph- (the connective -ko and plus siph-) and the inferential construction -ka siph- (the interrogative ender -ka plus siph-) are observed as early as the 15th C, when the Korean alphabet was rst created, while the connective siphi rst appeared in the late 19th C. Previous studies (H.H. Lee, 1994; Sohn, 1995; Ahn, 2005; Y.G. Lee, 2005) generally agree that the original meaning of siph- was to think to oneself, which allowed for the earlier dual functions of the desiderative and inferential meanings. Diachronic studies on siph- have thus far mostly focused on these auxiliary (desiderative and inferential) functions and their structural changes over time (Sohn, 1995; Ahn, 2005; Y.G. Lee, 2005). What is missing from these studies is an account of the mechanisms of semantic extension that led to the desiderative and inferential functions, and an account of the emergence of the connective siphi. Trying to ll the gap left by the earlier studies, the present study will investigate in detail the three paths of the diachronic development of siph-. In doing so, this study will also touch on the question of unidirectionality in grammaticalization. As shown in the denition by Hopper and Traugott (1993), at the core of grammaticalization theory is the unidirectionality hypothesis (e.g., a unidirectional change from a lexical to a grammatical category). In recent years, however, opponents of grammaticalization theory have compiled examples of degrammaticalization that appear to refute the unidirectionality hypothesis, and they argue that without an overarching directionality, linguistic changes classied under grammaticalization can be reduced to cases of reanalysis and analogical extension (Harris and Campbell, 1995; Newmeyer, 1998; Joseph, 2001). In defense of the unidirectionality hypothesis, proponents of grammaticalization theory have made efforts to sort out legitimate counterexamples from illegitimate ones and in doing so, have tried to distinguish grammaticalization from other related processes such as lateral conversion and lexicalization (Traugott, 2001, 2004; Lehmann, 2004). Among various approaches in this line of effort, in my view the most convincing and
The transliteration system adopted in this study is an extended Yale system developed by Seongha Rhee (1996) to suit the transliteration of Middle Korean (e.g., z for D). The abbreviations used in this article are ACC (accusative), CAUS (causality), COND (conditional), CONN (connective), COP (copular), DEC (declarative), DEF (deferential), FUT (future), GEN (genitive), HON (honoric), IMP (imperative), IND (indicative), INS (instrumental), INT (intransitive), LOC (locative), NF (non-nite), NOM (nominative), PASS (passive), PAST (past, old anterior), PERM (permission), PRES (present tense), PROG (progressive), RES (resultative), RETRO (retrospective), RL (relativizer), SUP (superlative), TOP (topic), TRA (transitive), and TRANS (transferentive).
2

1002

M. Kim / Journal of Pragmatics 42 (2010) 10001016

insightful argument is the one proposed by Kiparsky (forthcoming) and Lehmann (2004). They suggest that when examples of degrammaticalization (i.e., counterexamples to the unidirectionality hypothesis) emerge with the inuence of analogy to already existing constructions in the language, they may not be considered to be legitimate counterexamples to the unidirectionality hypothesis.3 First, Kiparsky (forthcoming) tries to unify grammaticalization with analogy (in contrast with Meillet (1912) who characterized them as two disjoint classes of change), and points out that grammaticalization is a kind of analogy. He explained that in contrast with traditional analogy, which is exemplar-based, grammaticalization involves constraintbased analogy that is driven by general principles and constraints of language (e.g., language-independent preferences grounded in universal grammar such as the fusion of two words into one for structural economy). Setting aside the question of whether I agree with this innovative denition of grammaticalization, his argument on degrammaticalization is illuminating: putative examples of degrammaticalization are developed by exemplar-based analogy, which is driven by language-specic constraints, and hence they are not cases of spontaneous degrammaticalization. In a similar vein, Lehmann (2004:19) also explains the role of analogy in some examples of degrammaticalization and distinguishes examples of analogically-oriented degrammaticalization from those of pure degrammaticalization. In the development of Korean siph-, I will argue that the emergence of the connective siphi is a case of analogically-oriented degrammaticalization. As will be shown later, the more concrete meaning of the connective siphi as or similar to developed from the more abstract inferential meaning of siph-. Although this development may appear to be a case of degrammaticalization, I will demonstrate that the emergence of siphi was inuenced by analogy to the existent constructions kathi and tusi due to the shared inferential meanings of kath-, tusha-, and siph-. In support of Kiparsky and Lehmann, the case of siphi will show that when one looks at just one development, it might appear to be an example of degrammaticalization, but examination of related grammatical markers in the language may demonstrate that the development was inuenced by analogy to existing constructions and hence may not be a legitimate example. The organization of the study is as follows. Section 2 will provide background information on the lexical source siph- as well as statistical information regarding the development of siph-. Sections 35 will examine the emergence of desiderative and inferential constructions with siph-. Section 6 will analyze the emergence of the connective siphi and section 7 will present the conclusion. 2. Lexical source and statistical information The original form of siph- in the 15th C was sikpu-; from the 16th C to the early 20th C, different transitional forms such as sipu-, sippu-, siphu-, and siphwu- co-existed. The morphological changes of sikpu- leading to siph- in Present Day Korean (hereafter PDK) can be roughly summarized as in (4). (4) sikpu- > 15th C sipu-/sippu- > 16-17th C sippu-/siphu- > 18-19th C siphPDK

Since the 15th C, siph- has been used as an auxiliary predicate and its occurrence has been limited to an auxiliary position that follows the main predicate in desiderative and inferential constructions; this pattern continues to the present. Two notes should be made about the lexical source siph- to think to oneself. First, siph- is a psychological predicate which is subject to the subjectivity constraint: it can only take rst person subjects in the present tense. As shown in the PDK examples (5b) and (6b), the use of third person subjects in the present tense without an additional marking (such as -e ha- do the action of) renders these sentences ungrammatical. (5) a. na-nun ku-kes-ul ha-ko siph-ta I-TOP that-thing-ACC do-ko siph-DEC I want to do that thing.

3 For example, in South Estonian, the abessive case sufx lta without has become a clitic, and this change from a case sufx to a clitic challenges the unidirectionality hypothesis. However, Kiparsky (forthcoming) explains that this development was in fact motivated by analogy to its antonym, the comitative clitic -ga with, which had arisen by grammaticalization from a lexical item.

M. Kim / Journal of Pragmatics 42 (2010) 10001016

1003

b. *ku-nun ku-kes-ul ha-ko siph-ta he-TOP that-thing-ACC do-ko siph-DEC He wants to do that thing. (6) a. na-nun i-kes-i cinswu-uy phalca-i-nka siph-ta I-TOP this-thing-NOM cinswu-GEN fate-COP-RL:PRES-ka siph-DEC (In my view), this probably is Cinswus fate.

b. *ku-nun i-kes-i cinswu-uy phalca-i-nka siph-ta he-TOP this-thing-NOM cinswu-GEN fate-COP-RL:PRES-ka siph-DEC (In his view), this probably is Cinswus fate. This subjectivity constraint was already observed in the 15th C. Compared to the English case (Postal, 1970), subjectivity constraints for psychological predicates are more widespread in Japanese and Korean. In Japanese, predicates expressing emotion (e.g., osoroshii to fear), desire (e.g., -tai to want to), and cognition (e.g., omou to think) are affected (Kuroda, 1973; Aoki, 1986), while in Korean, predicates expressing emotion (e.g., mwusep- to fear) and siph- to think to oneself are affected. Second, similar to most other Korean psychological predicates, siph- is an adjectival predicate (C.S. Suh, 1996). As in Japanese, adjectives in Korean are all predicates and hence are often called adjectival predicates, because they can be divided into stem and ending, and can take tense-aspect and modal markings. The fact that siph- is an adjectival predicate and not a verb is attested to by its conjugational pattern. As shown in (7a), while the present tense relativizer for verbs is nun, in (7b) siph- is combined with un, the present tense relativizer for adjectival predicates. (7) (verb mek- to eat plus nun) a. nay-ka kacang manhi mek-nun umsik-un lamyen-i-ta I-NOM SUP much eat-RL:PRES food-TOP ramen-COP-DEC The food that I eat most is ramen. (adjective siph- plus un) b. nay-ka kacang manhi mek-ko siph-un umsik-un lamyen-i-ta I-NOM SUP much eat-ko siph-RL:PRES food-TOP ramen-COP-DEC The food that I want to eat most is ramen. In short, the Korean siph- to think to oneself is a psychological, adjectival predicate and its attributes as such are retained in its grammaticalized uses in desiderative (e.g., V + ko siph-) and inferential (e.g., V + ka siph-) constructions. Table 1 describes the diachronic corpus data used for this study, which start with the 15th C when the Korean alphabet was rst created. The corpus data, which contain various genres of historical texts including religious texts, novels, and private letters, were made available by the National Academy of the Korean Language.
Table 1 Size of corpus for each century. Number of words 15th C 16th C 17th C 18th C 19th C and early 20th C Present Day Koreana Total
a

653,108 202,276 360,515 415,462 894,467 439,365 2,965,193

Various genres of written samples such as short novels and newspaper articles are included.

1004

M. Kim / Journal of Pragmatics 42 (2010) 10001016

Table 2 Diachronic change in frequency of the auxiliary adjective siph-. Frequency of siph-a 15th C 16th C 17th C 18th C 19th C and early 20th C Present Day Korean
a

2.8 18.3 13.0 26.5 66.7 137.2

The numbers are normalized and represent occurrences per 100,000 words.

Table 2 presents diachronic changes in the frequency of the predicate siph-. The table does not include tokens of the connective siphi because that did not appear until the late 19th C. Table 2 shows that the use of siph- has increased over time; an increase in frequency is typically expected in grammaticalizing constructions (see Bybee, 2003a, 2003b; Hopper and Traugott, 2003:126129). Also note from Table 2 that the frequency of siph- in the 15th C is low. This demonstrates that in the 15th C, the desiderative -ko siph- and the inferential -ka siph- are at an early stage of their grammaticalization, and hence reconstruction of their grammaticalization processes based on 15th C data is still possible. Of the desiderative and inferential functions of siph-, the rate of increase was much greater in the desiderative than in the inferential use. In the 15th C corpus, out of 2.8 occurrences of siph- per 100,000 words, 33% are desiderative tokens and 67% are inferential tokens, and this tendency for inferential tokens of siph- to be more frequent than desiderative ones continues until the 18th C. However, as the use of siph- started to increase drastically in the 19th and early 20th C, desiderative tokens began to outnumber inferential ones. For instance, in the PDK corpus, out of 137.2 occurrences of siph- per 100,000 words, 72% are desiderative and 28% are inferential, reversing the 15th C trend. The following sections will examine the historical development of grammatical constructions with siph-, starting with the desiderative case. 3. Desiderative constructions with siphTable 3 summarizes the four stages of diachronic change involving the desiderative -ko siph-, which I will examine in this section. In the 15th C, the original desiderative construction was -kocye sikpu-; see (10). In this rst stage, the meaning of desire was mostly carried by the connective -kocye, which signaled desiring or in desire of (Sohn, 1995); the following sikpu- (> siph-) only added the meaning of to think to oneself. As illustrated in examples (8) through (10), the desiderative constructions with -kocye in the 15th C were -kocye hA- (hA- to do, over 500 tokens), -kocye sAlanghA- (sAlanghA- to think, 18 tokens), and -kocye sikpu- (sikpu- to think to oneself, 6 tokens). Note that in -kocye sAlanghA-, the meaning of sAlanghA- is to think, similar to that of sikpu-.

Table 3 Development involving the desiderative -ko siph-. Form Stage 1 Stage 2 Stage 3 Stage 4 -kocye siphwant to -ko siphwant to -myen siphwould be great if -ko siph- > kophu want to Time Before 15th C 17th C 18th C Early 20th C Mechanism -kocye desiring plus siph- to think to oneself => desiderative construction Phonological reduction => siph- reanalyzed as to desire New form -myen if plus siph- to desire (through analogy with -myen tyoh-) Further phonological reduction

M. Kim / Journal of Pragmatics 42 (2010) 10001016

1005

Stage 1 (8)

ip-ey tyohAn mas mek-kocye mouth-LOC good taste eat-kocye mom-ay tyohAn os nip-kocye hA-l ss-i-la body-LOC good cloth wear-kocye do-RL DN-COP-DEC (The ve desires include) wanting to eat good tastes (food) for the mouth and wanting to wear good clothes for the body. (Seksang 1447, 3: 22b) 4 kanghan-aysye ka-kocye sAlanghA-nAn nakunay-ye Kanghan-LOC go-kocye think-RL traveler-VOC Traveler, who wants to go from Kanghan (area) (Twusi chokan 1481, 3: 40a) na-kocye sikpu-nye (Welinchenkang 1447, sang: 132) come.out-kocye siph-Q Do you want to come out?

(9)

(10)

In PDK, -kocye sAlanghA- has become obsolete with the semantic change of sAlanghA- (>salangha-) from to think to to love. -Kocye hA- is still in use in its modern form -koca ha-, but it underwent a semantic shift and denotes intention rather than desire (Sohn, 1995).5 In the 15th C, -kocye hA- was used more productively than -kocye sikpu-. However, with the psychological predicate sikpu-, -kocye sikpu- became a marker of desire while -kocye hA- with the verb hA- to do changed to mark intention. After -kocye sikpu- grammaticalized as a desiderative construction, in the 17th C, cye in -kocye began to be omitted, leading to the PDK form -ko siph-. Stage 2 (omission of cye starting in the 17th C) (11) al- ko sipu-o-ni (Chephay sine 1676, 5: 11a) know-ko siph-HON-CONN Because I want to know. . . Previous studies (Ahn, 2005; Y.G. Lee, 2005) have attributed the omission of cye to an overgeneralization of the periphrastic constructions of connective ko and plus auxiliary verb. It is true that such a trend was prevalent at the time because of the explosive expansion of the connective -ko and. From the 15th C until the end of the 17th C, the use of the connective -ko increased drastically (by 92%; M. Kim, forthcoming). At the same time, the omission of cye was also made possible because with its grammaticalization as a desiderative construction, -kocye sikpu- had undergone semantic reanalysis and transfer, as shown in (12). With the frequent use of -kocye sikpu- as a single construction, the reduction from -kocye sikpu- to -ko sikpu- did not interfere with its recognition. In addition, because the use of sikpu- had been limited to desiderative or inferential constructions, speakers could easily associate the meaning of desire with sikpu-. These two factors contributed to the reduction of -kocye to -ko. Even at present, most speakers of Korean would construe siph- in -ko siph- as the main semantic carrier that expresses desire, ignorant of its original lexical meaning. (12) [[propositional [[propositional [[propositional [[propositional content -kocye*] sikpu-] content] [-kocye* sikpu-]] content] [-kocye* sikpu-*]] content] [-ko sikpu-*]]

(*main semantic carrier of desire)

In citing the sources of the examples, I follow the conventions used in Korean studies. Each example from historical documents will be followed by a mark such as (Sekposangcel 1447, 11:37b). Sekposangcel is the title of the cited text; the number 1447 represents the year of publication; 11 represents the volume number; and 37b indicates the page number. 5 The example below demonstrates that in the 18th C, -ko siph- marked desire while -kocye ha- marked intention. aloy-ko sip-un il-ul alloy-kocye hA-toy tell-ko sip-RL thing-ACC tell-kocye do-CIRCUM He intended to tell the things that he wanted to tell (to his superior) (Myengulok 1777, sang: 34b)

1006

M. Kim / Journal of Pragmatics 42 (2010) 10001016

In contrast to the case of -kocye sikpu-, the 15th C -kocye hA- construction continued its form into PDK (-koca ha-) without going through the omission of cye. The verb ha- to do is productively used and has various secondary meanings (e.g., to say, pro-verb to do). Hence, the reassignment of the desiderative meaning to ha- is not feasible with -kocye ha-; in fact, the reduced form -ko ha- becomes a quotative construction -ko ha- (the quotative particle -ko plus the say verb ha-). The contrasting examples of -kocye sikpu- and -kocye ha- support the argument that the phonological reduction from -kocye sikpu- to -ko sikpu- was facilitated by semantic reanalysis of the construction, as shown in (12). Starting in the 18th C, the new borrowed desiderative meaning of siph- enabled it to combine with the conditional myen if creating a new desiderative construction, -myen siph-. With the replacement of the connective -ko by the conditional marker -myen, the construction -myen siph- signies it would be great if. . . ; see (13). Stage 3 (emergence of a new construction, -myen sipu-, in the 18th C) (13) nathana-key hA-si-myen sipu-te-nila (Hancwunglok 18th C) appear-NF do-HON-COND siph-RETRO-DEC I wished he could make it appear. . . It should be noted that the development of the new construction -myen siph- was prompted by analogy to the existing desiderative construction -myen tyoh- it would be great if. . . (its PDK form is -myen coh-). More specically, based on the combinatoric pattern of -myen tyoh- (-myen if and the adjective tyoh- to be good), the conditional myen and the desiderative element siph- were put together, resulting in a similar construction -myen siph- it would be great if. . . ; see (14). (14) swumi o-myen tyoh-kenmanAn (Kim family letters 16th C) Swumi come-COND good-CONC It would be great if Swumi (persons name) can come but. . .

In the 15th C, although the use of -myen was already very productive, its occurrence was limited to typical conditional sentences, if X happens then Y happens. Starting in the 16th C, however, -myen tyoh- began to express the desiderative meaning, it would be great if X happens.6 Then, after the grammaticalization of the desiderative construction -ko siph-, another desiderative construction with the conditional -myen, -myen siph-, arose in the 18th C, modeled after -myen tyoh-; see (15). (15) [[propositional content] [-myen tyoh-]] -myen if tyoh- to be good (16th C) it would be great if. . . [[propositional content] [-ko siph-]] -ko siph- want to (17th C) [[propositional content] [-myen siph-]] through analogy with -myen tyoh- (18th C) it would be great if. . .

Kiparsky (forthcoming) points out that because morphological categories are compositional, new grammatical categories can be created by analogy, making use of existing combinatoric patterns in the language. (15) shows that emergence of -myen siph- could have been facilitated by the existing desiderative construction -myen tyoh- in a similar fashion. Finally, with the popular use of -ko siph-, another round of phonological reduction to -kophu- was observed in early 20th C. Today, -kophu- is not commonly used but is still recognized as a contracted variation of -ko siph-; see (16). Stage 4 (further reduction from -ko siph- to kophu) (16) tAlk-uy al-ul mAntAl-a mek-koph(u)-Ana chicken-GEN egg-ACC make-CONN eat-kophu-CONS Although we want to (articially) make and eat eggs . . . (Kyenghyang pokam 19061910)
Although the collocation of -myen and tyoh- is observed even in the 15th C, it expresses a conditional meaning and not a desiderative meaning. For instance, it is extensively used in the medical book Kwukup kani (1489) for the meaning if you do X, you will feel better.
6

M. Kim / Journal of Pragmatics 42 (2010) 10001016

1007

To summarize, the development of the desiderative -ko siph- is a typical case of grammaticalization that shows semantic reanalysis and phonological reduction. The compositional meaning of the connective -kocye desiring and the psychological predicate siph- to think to oneself could easily induce the conversational implicature of the desiderative meaning. Over use and time, this implicature of -kocye siph- became generalized (Levinson, 2000) or conventionalized (Traugott and Dasher, 2002; Hopper and Traugott, 2003), which eventually led to its grammaticalization as a desiderative construction. Its grammaticalization also enabled the emergence of an additional desiderative construction, -myen siph-, by analogy to -myen tyoh-. The development of inferential constructions with siph-, which will be discussed next, shows a similar pattern. After -ka siph-, the rst inferential construction with siph-, was grammaticalized, additional grammatical constructions that make use of the inferential siph- (e.g., -tus siph- and -seng siph-) could be subsequently created through partial replacement of a morpheme. 4. Inferential constructions with siphTable 4 summarizes the ve stages of diachronic change that resulted in the three inferential markers, -ka siph-, -tus siph-, and -seng siph-. Heine and Kuteva (2002) show that verbs marking cognition (e.g., to think) and utterance (e.g., to say) tend to show similar patterns of development across different languages. The case of siph- to think to oneself also attests to the close relationship of these two types of verbs: the complement clause of siph- takes the form of reported speech and as such, its propositional content is formatted as what the speaker says to himself; see (17). Stage 1. Lexical Stage: to think (to oneself) (17) yeksim-i na-ni twu-ela essti-li sipu-ni dislike-NOM emerge-CONN leave-IMP how.do-FUT siph-CONN Annoyance starts to rise (about your mothers jealousy toward concubines), so I think (to myself) leave her alone, what can she do? (Kim family letters 16th C, 83) In historical texts, inferential siph- is almost always combined with an interrogative complement clause that ends in the interrogative ender -ka.7 As in (17), when siph- is combined with wh-questions, the reading of siph- is close to its original lexical meaning to think to oneself (stage 1). On the other hand, when siph- is combined with yesno questions, its probability inference can emerge, suggesting that the combination think to oneself/wonder whether it is. . . had led to the pragmatic inference of probability (stage 2); see (18) through (20).
Table 4 Development involving the inferential -ka siph-. Form Stage 1 Stage 2 Stage 3 Stage 4 Stage 5 -ka siphthink whether it is. -ka siphprobability -ka siphhigh probability -tus siphhigh probability -seng siphhigh probability Time Before 15th C Before 15th C 17th C 18th C Early 20th C Mechanism Lexical stage: -ka interrogative ender plus siph- to think to oneself Grammaticalization into an inferential marker probably Capable of marking high probability (transitional phase before the emergence of -tus siph-) Through analogy with the inferential marker tus haThrough analogy between dependent nouns tus and seng

In PDK, there are two question markers, -ka and -kka, but their original form in the 15th C was the same, ka.

1008

M. Kim / Journal of Pragmatics 42 (2010) 10001016

Stage 2. Probability (context: after failing tests to become a government ofcer, a man writes to his wife) (18) kul-to nAm-man mot-hA-keniwa phalcA-y kule-n-ka sipu-ni Writing-also other-like NEG-do-CONN fate-NOM so.do-RL:PRES-ka siph-CONN My writing is also not on a par with the others and, I think to myself, is this my fate? = > probably this is my fate. (Kim family letters 16th C, 51) (context: a servant implies that the queen had sympathized with the traitors.) (19) kul-no tewuk syelwehA-osi-nAn-ka siputa hAn-i That-INS more sorrowful-HON-RL:PRES-ka siph-DEC say-CONN He said (to the King), I think to myself is the Queen more sorrowful because of that (the failed plan to make her son King)? => He said (to the King) that the Queen is probably more sorrowful because of that. (Kyeychwuk ilki 17th C, sang: 14a) (20) tasi poy-o-l ka sipu-ci anihA-ya again see-HUM-RL:FUT ka-siph-COMP NEG-CONN Since I probably will not see him again. . . (Hancwunglok 18th C, 233)

Yesno questions logically indicate a 50% possibility, but in real life, the very act of raising and pondering the question allows the inference of a higher probability. For instance, in (18), after failing many tests to become a government ofcer, a man writes a letter to his wife saying that he wonders whether the reason that his writing is not on a par with that of the others and that he has failed the tests is because of his fate. Of all the possible reasons for his failure, the husband raises the possibility of fate, and thus, though a yesno question is employed, the message that the addressee actually computes is a qualied statement of the mans subjective thinking probably it is my fate that I keep failing at the tests. Similarly, in (19), a court servant who has conspired against the queen is telling the king that he wonders whether the reason why the queen is sorrowful is that the traitors plot to replace the king with her son has failed. By using a yesno question plus siph-, the propositional content is presented as his personal cause for wonder, but by mentioning the failed plot out of other possible reasons for the queens sorrow, the speaker also expresses that in his view, the failed plot is the most probable answer. In this way, the combination of a yesno question plus siph- I think to myself whether it is . . . can invite the addressee to compute a probability meaning; through frequent use, this generalized pragmatic inference could be grammaticalized as a grammatical function of -ka siph-. Finally, (20) is a clear example of a probability use. The literal meaning of the negative sentence (20) should be I dont wonder whether I will see him again because the negative marker -ci anihA- is sufxed to siph-. However, with the grammaticalization of -ka siph- as a single inferential construction, the actual meaning delivered here is the negative probability I probably will not see him again. In a similar vein, note that with the grammaticalization of -ka siph- as an inferential construction, the original interrogative meaning expressed by -ka (i.e., V + interrogative ender ka + siph-) has more or less eroded. Eventually, in the 19th C (e.g., Hancwunglok 19th C, page 250), we can nd examples in which inferential siph- is combined with the V in a declarative ending (i.e., V + declarative ender ta + siph-); see (36). After -ka siph- grammaticalized as an inferential construction, it could expand this function to express an even higher probability, and thus could imply prediction; see (21) and (22). Stage 3. Probably. . . => in Present Day Korean replaced by tus siph(21) nAyil-un chenki tyohA-l-ka sipu-ta yekuy salAm-to nilu-o-ni tomorrow-TOP weather good-RL:FUT-ka siph-DEC here people-also talk-HON-CONN Since people here say that the weather will be good tomorrow (Chephay sine 1676, 6: 13b) (22) i pAy-nAn pwunmyengi hwanthAy syen-i-n-ka sipuoy-ta this ship-TOP certainly returning ship-COP-RL:PRES-ka siph-HUM-DEC This ship would be certainly a returning ship. (Kayswuchephay 1748, 10cwung: 19b)

M. Kim / Journal of Pragmatics 42 (2010) 10001016

1009

In (21), when Korean envoys are getting ready to sail back home from Japan, the Japanese locals collectively make a prediction about the weather to decide upon the departure date; here -ka siph- marks a high probability that is based on collective judgment. In (22) as well, as shown by the use of the adverb pwunmyengi certainly, -ka siph- marks high probability. It should be noted that sentences like (21) and (22) sound awkward in PDK and are found only in historical texts, mostly in the 17th and 18th C; in PDK, propositions with such high epistemic certainty are expressed by another inferential construction, -tus siph-. The grammaticalization of the inferential -ka siph- prompted the emergence of -tus siph- in the 18th C. The interrogative marker -ka in -ka siph- was replaced by the dependent noun tus similarity and this brought about the new construction -tus siph-. Compared to -ka siph-, which contains the interrogative ender -ka, -tus siph- with the similarity component tus signies greater epistemic certainty. Stage 4. Probability (emergence of the new form tus sipu- in the 18th C) (23) icey yekuy eps-uni naka-ntus sipwu-ta Now here not.exist-CONN go.out-RL:PAST-tus siph-DEC Now (the young man) is not here, it seems that he has gone outside. (Chenge nokeltay 1765, 5: 5b) The emergence of -tus siph- was motivated by analogy to another inferential marker, tus ha-. In the 15th C, as will be further discussed in section 6.2, the adjectival predicate tusha- (made of tus similarity plus ha- to do) already functioned as a probability marker, and the same dependent noun tus is used in tusha- as in -tus siph-. After the grammaticalization of -ka siph- as a probability construction, the auxiliary predicate siph- (rather than the interrogative ender -ka) seems to have been reanalyzed as the main carrier of the probability meaning. As in the case of the desiderative -ko siph-, this semantic reanalysis was further facilitated by the fact that occurrences of siph- had been limited to desiderative and inferential constructions. Then, by replacing the interrogative ender -ka with the dependent noun tus marking similarity, the new inferential construction -tus siph- which expresses higher probability emerged; see (24). (24)[[propositional content -ka] siph-] interrogative ender -ka plus siph- to think to oneself [[Propositional content] [-ka siph-]] interrogative ender -ka plus siph- to be probable [[Propositional content] [-tus siph-]] tus similarity plus siph- to be probable (based on analogy with the inferential tusha) [[Propositional content] [-seng siph-]] seng appearance plus siph- to be probable (based on analogy between tus and seng) Finally, as shown in (24), following the precedent of -tus siph-, another inferential construction -seng siph- emerged in the early 20th C. This construction arose from the replacement of the dependent noun tus with another dependent noun seng (also sang) appearance, similarity, a paradigmatic replacement that was motivated by the analogy between the two dependent nouns tus and seng, which share similar semantics.8 Through this process, -seng siphemerged; similar to -tus siph-, it signies high probability as shown in (25). Stage 5. (emergence of -seng siph- in early 20th C) (25) kacok-i wa-(a)sye elengteleng hA-ki-man-hA-yto family-NOM come-CONN bustle do-NMZ-only-do-CONC com nah-ulsang sip-e a little better-RL:FUT-sang siph-CONN If the family comes and makes a bustle, even that will probably make him feel better (Kumkangmwun 1914, 27)
8 Not just the meaning of similarity (see section 6.2.) but also the meaning of appearance (e.g., seng appearance in seng siph-) can bring about the meaning of probability. For instance, in Korean, moyang-i-ta be/have the look of and in Japanese yoo da be/have the look of can signify probability.

1010

M. Kim / Journal of Pragmatics 42 (2010) 10001016

To summarize, as a result of these multiple rounds of development, in PDK, three constructions with siph- -ka siph-, -tus siph-, and -seng siph- share the inferential function. The development of inferential -ka siph- is a typical case of grammaticalization: a reanalysis took place within its associative use, and a conventionalized pragmatic inference became a grammatical function. After its emergence, -tus siph- and -seng siph- were subsequently created by the partial replacement of -ka siph-. Similar to the development of desiderative constructions with siph- discussed in section 3, section 4 demonstrates again that in Korean partial replacement of morphemes based on existing constructions can productively induce additional grammatical markers, enriching the repertoire of the functional domain. 5. Development from content meaning to procedural meaning The inferential construction -ka siph- developed making use of siph- to think to oneself. The development of periphrastic I think constructions into epistemic markers has been observed in other languages besides Korean, and interestingly, they have all grammaticalized into markers with a similar position on the epistemic scale, probably; see (26). (26) n, 1991); Tok Pisin ating English I think as in I think shes home. (Thompson and Mulac, 1991; Givo (I think) > maybe, probably (Keesing, 1991; Brinton and Traugott, 2005)

On the basis of Sweetser (1990) and other evidence, Traugott (2002) proposes a unidirectional tendency in semantic changes including those that lead to the development of epistemic meanings as in (27). The proposition in (27d) that sociophysical meanings can engender epistemic meanings but not vice versa can actually be subsumed under Traugotts famous argument that semantic change moves from content meaning to procedural meaning and not vice versa (27c), which is also in accordance with the unidirectionality hypothesis of grammaticalization, predicting changes from more concrete to more abstract meanings and not vice versa. Content meanings can be understood as meanings that describe the sociophysical world and concern the propositions themselves. On the other hand, according to Traugott, procedural meanings index metatextual relations between propositions or between propositions and the non-linguistic context; hence they can cue addressees to speakers and writers attitudes to the discourse and the participants in it (Traugott, 2002:33). (27) a. b. c. d. SPACE > TIME not vice versa (Traugott, 2002:32) DEONTIC > EPISTEMIC not vice versa CONTENT > PROCEDURAL not vice versa SOCIOPHYSICAL > EPISTEMIC > METATEXTUAL not vice versa

The emergence of the Korean inferential -ka siph- ts Traugotts model well, in that the content meaning of -ka siph- to think to oneself whether it is. . . developed into an epistemic, procedural meaning. The next step of development, which led to the emergence of the connective siphi, however, shows some discrepancies from what has been proposed in (27c). Section 6 will discuss the development of the connective siphi and demonstrate that the content meaning of similarity of the connective siphi could emerge from the procedural meaning of inference of the auxiliary predicate siph-. 6. The clausal connective siphi In the late 19th C, a new construction siphi emerged, derived from the combination of the auxiliary, adjectival predicate siph- and the adverbial sufx i. Adverbs often function as conjunctions that combine clauses in different languages (Brinton and Traugott, 2005:132136). Attached to verbs with the declarative ender -ta, V-ta siphi also functions as a clausal connective, meaning as or similar to, and combines the subordinate clause marking manner with the main clause that follows. In discourse, V-ta siphi is used for two distinct functions. First, as in (28), when used with a set of limited verbs such as po- to see and al- to know, V-ta siphi meaning as marks evidential grounds (e.g., as you see, as I said) and is used in speech acts seeking the hearers understanding. Second, as in (29), the simile construction V-ta siphi meaning similar to V or like V-ing modies the following verb (e.g., run like ying).

M. Kim / Journal of Pragmatics 42 (2010) 10001016

1011

(28)

(29)

ney-ka al-ta-siphi na-nun acik haksayng-i-ta. you-TOP know-ta-siphi I-TOP yet student-COP-DEC As you know, I am still a student. na-nun keuy nal-ta-siphi ttwi-ess-ta I-TOP almost y-ta-siphi run-PAST-DEC I ran almost like ying (i.e., I ran that fast).

Among previous studies, Y.G. Lee (2005) notes that the connective siphi appears to have derived from the auxiliary function of siph-. Evidence from my corpus substantiates this derivation. In PDK orthography, the connective siphi is written as si.phi (. indicates the syllabic boundary). However, when this construction rst emerged in the late 19th C, it was written as sip.hi, sip.i or si.phi. The three varied forms of the connective siphi can be traced to the three corresponding forms of the auxiliary predicate siph which were used at that time: as shown below, sip.hi is derived from the predicate form sip.hu- (see (30) and (31)), sip.i is derived from sip.u- (see (32) and (33)), and si.phi is derived from si.phwu- (see (34) and (35)). Also, in the 19th C and early 20th C corpus, the order in terms of decreasing frequency among the three connective variations was sip.hi (49 tokens), sip.i (3 tokens), and then si.phi (1 token). This order corresponds to the frequency order of the predicate variations in the same corpus, as sip.hu- (269 tokens) is most frequent, followed by sip.u- (196 tokens), and si.phwu- (82 tokens). (30) tocyuhA-ta sip.hi stena-s-ta-(ha)-nAntAy run.away-ta sip.hi leave-PAST-DEC-(say)-CIRCUM It is said that (she) left (the town) as if running away. . . (Pinsangsel early 20th C, 955) kwunsyu-eykey hAy-lul tanghA-n tus sip.h-uni governor-DAT harm-ACC suffer-RL:PAST tus sip.h-CONN (My husband) is probably harmed by the county governor. . . (Sey kem ceng early 20th C, 127) lyengkam-to a-si-ta sip.i cyoyonghA-ko amo-to eps-ini Old.man-also know-HON-ta sip.i quiet-CONN anybody-even not.exist-CONN As you (old man) know, (my house) is quiet and nobody is there. . . (Hanwel early 20th C) mischye tAyka-lka sip.u-ci anh-ta quite arrive.in.time-RL:FUT-ka sip.u-COMP NEG-DEC (We) probably would not quite arrive there in time. (Pihayngsen early 20th C) ton hAn phwun epsi pil-e mek-ta si.phi hA-ko money one phwun without beg-CONN eat-ta si.phi do-CONN Without one phwun (old money classier), did like begging and eating (lived like a beggar) and . . . (Independence paper 1896) ep-nAn il-ul cyangmAnhA-n tus-to si.phwu-ko not.exist-RL thing-ACC make-RL tus-even si.phwu-CONN It also seems possible that he made up things/stories that did not happen. (Kwi-uy-seng 1906)

(31)

(32)

(33)

(34)

(35)

It should be noted that, as shown in examples (30) through (35), when the various early forms of siphi rst emerged, the derivational relationship between predicate and connective was traceable in their orthographies because their rst syllables had the same form. In contrast, PDK orthography has partially eroded the forms derivational relationship: for the predicate, siph- is used, and for the connective, si.phi is used instead of siph.i. In PDK orthography, it is common practice to use the syllable boundary to keep the derivational sufx i separate from the host to which it is attached. However, in the case of si.phi, the sufx i has coalesced with the predicate host resulting in the use of si.phi instead of siph.i. In spite of their derivational relationship, the differentiation in orthography between the auxiliary predicate and connective uses attests that speakers of Korean perceive some disjunction between the two uses and do not consider this to be a seamless derivation. The next section addresses this disjunction.

1012

M. Kim / Journal of Pragmatics 42 (2010) 10001016

6.1. From inferential siph- to connective siphi In mapping out the developmental connection between the auxiliary predicate siph- and the connective siphi, it seems that the inferential siph- rather than the desiderative siph- was involved. One reason is that a semantic connection can be drawn between the similarity meaning and the inferential meaning; section 6.2 will show that across different languages, the similarity meaning often develops into the probability meaning. A syntactic change in the use of the inferential siph- also points to its genealogical tie with the connective siphi. As mentioned in section 4, in the 19th C, the interrogative ender -ka in the inferential construction -ka siph- could be replaced by the declarative ender -ta; see (36). The inferential construction V-ta siph- seems specically responsible for the development of the connective construction V-ta siphi. (36) cyuk-ul kos-ey spacy-es-ta siphwu-n sAyngkak-spwun-i-la (Kwi-uy-seng 1906) die-RL place-LOC fall-PAST-ta siph-RL thought-only-COP-DEC The only thought that (I) have is that (I) seem to have fallen in the death place.

In sum, of the desiderative and inferential meanings of siph-, the connective siphi seems to have derived from the inferential rather than from the desiderative meaning. However, this development from the inferential siph- to the connective siphi is difcult to explain using the usual parameters of grammaticalization theory. There is a strong developmental tendency for similarity meanings to give rise to probability meanings, which supports the unidirectionality hypothesis presented in (27c, d). On the other hand, the semantic change from the probability construction V-ta siph- to the similarity construction V-ta siphi is a change from a more abstract procedural meaning to a more concrete content meaning, which appears to go against the unidirectionality hypothesis. The development of siphi is motivated by the sufxing of the adverbial derivation marker i. Brinton and Traugott (2005:132136) examine the derivation of adverbs with the addition of -ly in English and propose that these derivations often result in more abstract meanings and even increased grammaticality. For example, almost all degree adverbs with -ly (e.g., highly and really) are derived from more literal meanings. The emergence of the probability meaning of likely from like also seems to be a case of derivation with -ly that came to assume a more grammatical function. I am not arguing that Korean adverbs that are derived by the sufxing of -i have become more grammatical as in the English cases with -ly. But I am pointing out that the development of the similarity meaning of siphi from the derivation of the inferential siph cannot be well explained by the parameters of grammaticalization and even seems to be a counterexample to the unidirectionality hypothesis. In the next section, I will show that these discrepancies occurred because the emergence of siphi was made possible by the inuence of analogy to already existing Korean constructions with similar functions. Specically, as presented in Fig. 1, it emerged by analogy to the adverbials kathi and tusi. These are derived from kath- and tus(ha)-, which denote both similarity and probability and as such share the inferential meaning with siph-. 6.2. Development from similarity to probability There is a strong tendency for words meaning similarity to obtain epistemic meanings, as witnessed in the cases of like (likely) and seem in English, and of the two adjectival predicates kath- and tusha- in Korean. First, the adjective

Fig. 1. Emergence of siphi by analogy with tusi and kathi.

M. Kim / Journal of Pragmatics 42 (2010) 10001016

1013

like stemmed from the Old English gelic which denotes of the same form or shape; derived from this adjectival meaning of like, likely (<Old English geliclic) came to mean probability as shown in (37).9 (37) (38) a. a. b. a cup of water and a like amount of our The likely cause of his death. It is likely to rain today. John seems taller than Jane. It seems he was here. I seem to have lost it. (similarity) (probability) (similarity) (probability)

The case of seem also attests to this common path of development. The origin of seem can be traced back to Old Norse sma to conform to and is also related to Old English seman to bring to agreement. As such, it shares an Indo-European base with the adjectives same and similar. Presently, it denotes the meaning of both to have the look of and probably; see (38). In Korean, both kath- and tusha- followed a similar track. Since at least the 15th C, kath- (< kAthA-) had been used to denote similarity. Later, around the 17th C, probability examples of kath- started to appear. In PDK, kath- carries both similarity and probability meanings; see (39) and (40). (39) si-cong-i kAthA-si-lssAy rst-end-NOM kath-HON-CONN since the beginning and the ending are the same.. (Yongpiechenka 1447, 79) twue kwicyel-i tusAlo cwucakhA-n kes kAthA-n-di-la about.two sentences-NOM oneself fabricate-RL:PAST DN kath-RL-COMP-CONN It seemed that he fabricated two or three sentences (in the writing) by himself. . . (Chenuy sokam 1755, 4: 78b)

(40)

Similar to kath-, tusha- (< tAshA-) was used to express similarity in the 15th C, and it also already carried a probability meaning at that time; see (41) and (42). The adjectival predicate tusha- is composed of the dependent noun tus similarity and ha- to do, and the tus in tusha- is the same dependent noun as tus in -tus siph-. (41) pam etuwun cwung-ey khum hwayspul hye-n tAshA-mye night dark middle-LOC big torch light-RL:PRES tus ha-CONN It (Buddhas presence) was like lighting up a torch in the middle of the dark night. . . (Pephwa 1463, 4: 138a) ney pap-i cyak-ul tAshA-kona your meal-NOM small-RL:FUT tus ha-DEC (Penyek nokeltay 1517, sang: 40b) Your meal will probably be small for you (probably not be enough for you)

(42)

With respect to the dual functions of kath- and tusha-, what is most relevant to the current study is that the two adjectival predicates kath- and tusha- gave rise to the adverbials kathi and tusi, creating a model for the auxiliary predicate siph- to give rise to siphi. 6.3. The emergence of siphi by analogy to kathi and tusi As just discussed, kath- and tusha- expressed similarity as early as the 15th C; the adverbial forms kathi and tusi were also already in use at that time. They denoted the meaning as or similar to; see (43) and (44). Kathi (< kAthi) arose from the sufxing of the adverbial sufx i to the adjectival predicate kath- (< kAthA-). In the case of tusi (< tAsi), although it is clear that it was derived by the sufxing of i, it cannot easily be determined whether its host form was the dependent noun tus (< tAs) or the adjectival predicate tusha- (< tAshA-). In the 15th and 16th C, the deletion of ha- was common (K.K. Lee, 1987); hence, the use of tus in place of tusha- was also common (e.g., the use of tAs-ketun instead of tAshA-ketun in Kwukup kani 1489, 1: 20a). Because tus and tusha- both denote similarity, henceforth I will simply note the origin of tusi as tus(ha)-.
9

In this study, etymological explanations are taken from Websters New World Dictionary (1972).

1014

M. Kim / Journal of Pragmatics 42 (2010) 10001016

(43)

(44)

puthye kAlAchi-sy-am kAthi hA-zA-wa i pep-ul nepi phye-li-la Buddha teach-HON-NMZ kathi do-HUM-CONN this law-ACC widely spread-FUT-DEC We will do as Buddha taught, and spread his law widely. (Pephwa 1463, 4: 192a) chang-Ay tuly-enAl cwuk-un tAsi ca-taka window-LOC enter-CONN die-RL tAsi sleep-TRANS Since (the sunshine) came into the window, I who slept like being dead. . . (Twusi chokan 1481, 22: 1a)

Morphologically, the three constructions kathi, tusi, and siphi are all derived from the sufxing of the adverbial sufx i; semantically, they all have the meaning of as or similar to.10 Returning to the question of the emergence of siphi, on the basis of formal and functional similarities, I suggest that the development of siphi was inuenced by analogy to kathi and tusi and that this analogical connection was made because of the common probability functions of kath-, tus(ha)-, and siph-. Through frequent use in the inferential constructions of -ka siph- and -tus siph-, the common factor siph- was recognized as a marker of probability. Then, the functional overlap among the three probability markers, kath-, tus(ha)-, and siph- appears to have triggered the emergence of siphi from inferential siph- by analogy to the precedent pairs of kath- and kathi, and tus(ha)- and tusi. What should be noted is that while kathi and tusi emerged from the similarity meanings of kath- and tus(ha)-, siphi, which appeared much later in the late 19th C, emerged from the probability meaning of siph-. The development of similarity into probability is cross-linguistically common whereas the development of probability into similarity is not. The connective siphi emerged through this uncommon route with inuence from the existing examples of kathi and tusi. The emergence of siphi could appear to be a case of degrammaticalization. However, its development was inuenced by analogy to existent constructions, and hence should be classied as analogically-oriented degrammaticalization (Lehmann, 2004) and not as a case of legitimate, spontaneous degrammaticalization. 7. Conclusion Using diachronic corpus data, I have analyzed the diachronic development of various grammatical markers with siph- in Korean. I have shown that once the desiderative -ko siph- and the inferential -ka siph- were grammaticalized through the conventionalization of pragmatic inferences within their usual contexts of use, additional desiderative and inferential constructions (e.g., -myen siph-, -tus siph-, and -seng siph-) could be created through analogy, which made use of the core semantic component siph-. In particular, the emergence of various desiderative (-ko siph- and -myen siph-) and inferential (-ka siph-, -tus siph-, and -seng siph-) markers with the shared component siph- demonstrates two characteristics that contribute to the productivity of analogy in creating new grammatical markers. First, due to the compositional nature of morphemes, new constructions can be created by partial replacement (e.g., of one or two morphemes) of a construction modeled after the compositional pattern of existing constructions with similar functions (Kiparsky, forthcoming). Second, a pattern of development motivated by analogy can be easily duplicated by functionally similar grammatical markers. This study has also touched on issues regarding the unidirectionality hypothesis of the grammaticalization theory. In defense of the unidirectionality hypothesis, proponents of the grammaticalization theory have tried to distinguish grammaticalization from other related phenomena (such as lateral conversion and lexicalization) and to distinguish illegitimate counterexamples from legitimate ones. Recently, Traugott (2001, 2004) has accepted that there are legitimate counterexamples, specically, those of exaptation (Lass, 1990, 1997; Vincent, 1995; Norde, 2001). Exaptation refers to a functional renewal through which a marginal grammatical marker at the last stage of grammaticalization develops into a more productive morpheme with a new function (e.g., the reanalysis of the perfect vs. aorist distinction in Indo-European as the singular vs. plural past distinction in Germanic; Lass, 1997:317). Subsequently, in their 2003 revision of the book Grammaticalization, Hopper and Traugott signicantly changed the
Compared to siphi, which is always used in the form of Verb-ta siphi, kathi and tusi can be combined with more diverse constructions. Kathi can be combined with nouns and nominalized clauses, with or without comitative markers such as wa, hako, and lang. Tusi can be combined with any predicates or clauses ending with relativizers since tusi contains the dependent noun tus. For this reason, I will label siphi as a connective, and kathi and tusi as adverbials.
10

M. Kim / Journal of Pragmatics 42 (2010) 10001016

1015

denition of grammaticalization (see note 1 of this article). At the same time, Hopper and Traugott (2003:232) emphasize that while examples that follow the unidirectionality hypothesis show robust tendencies, counterexamples to grammaticalization are sporadic and hence can be considered to be exceptions. In line with discussions on illegitimate counterexamples to the unidirectionality hypothesis, I have examined cases involving analogy with the Korean example of siphi. Supporting the proposals of Kiparsky (forthcoming) and Lehmann (2004), I have argued in this study that examples of degrammaticalization that are inuenced by analogy (analogically-oriented degrammaticalization) may not be legitimate counterexamples. Their irregular directionality is derived from the inuence of analogy, which is driven by prominent examples specic to a given language; as such its directionality cannot be easily generalized across languages. Acknowledgements An earlier version of this article was presented at the 16th Japanese Korean Linguistics conference at Kyoto University, Japan in 2006. I would like to thank the conference participants, especially Professors Chungmin Lee, Shoichi Iwasaki, Sung-Ock Sohn, John Whitman, and Alan Kim for their insightful comments. I also sincerely thank the two anonymous reviewers and the editor. Any remaining errors are my own. References
Ahn, Joo-hoh, 2005. Siphta kwumwunuy thongsicek kochal (Diachronic study on siphta phrases). Emwunhak 90, 7396. Aoki, Haruo, 1986. Evidentials in Japanese. In: Chafe, W., Nichols, J. (Eds.), Evidentiality: The Linguistic Coding of Epistemology. Ablex, Norwood, NJ, pp. 223238. Brinton, Laurel, Traugott, Elizabeth C., 2005. Lexicalization and Language Change. Cambridge University Press, Cambridge. Bybee, Joan L., 2003a. Cognitive processes in grammaticalization. In: Tomasello, M. (Ed.), The New Psychology of Language, vol. II. Erlbaum, Mahwah, NJ, pp. 145167. Bybee, Joan L., 2003b. Mechanisms of change in grammaticization: the role of frequency. In: Janda, R., Joseph, B. (Eds.), Handbook of Historical Linguistics. Blackwell, Oxford, pp. 602623. n, Talmy, 1991. Serial verbs and the mental reality of event: grammatical vs. cognitive packaging. In: Traugott, E., Heine, B. (Eds.), Givo Approaches to Grammaticalization, vol. I. John Benjamins, Amsterdam, pp. 81127. Harris, Alice, Campbell, Lyle, 1995. Historical Syntax in Cross-Linguistic Perspective. Cambridge University Press, Cambridge. Heine, Bernd, Kuteva, Tania, 2002. World Lexicon of Grammaticalization. Cambridge University Press, Cambridge. Hopper, Paul, Traugott, Elizabeth C., 2003. Grammaticalization. Cambridge University Press, Cambridge, First published in 1993. Joseph, Brian, 2001. Is there such a thing as grammaticalization? Language Sciences 22, 265340. Keesing, Roger M., 1991. Substrates, calquing and grammaticalization in Melanesian Pidgin. In: Traugott, E., Heine, B. (Eds.), Approaches to Grammaticalization, vol. I. John Benjamins, Amsterdam, pp. 315342. Kim, Minju, forthcoming. Grammaticalization in Korean: the Evolution of the Existential Verb. Saffron Books, Eastern Art Publishing, London. Kiparsky, Paul, forthcoming. Grammaticalization as optimization. In: Jonas, D. (Ed.), DIGS VIII Proceedings. Oxford University Press, Oxford. Kuroda, Susumu, 1973. Where epistemology, style and grammar meet: a case study from Japanese. In: Anderson, R., Kiparsky, P. (Eds.), A Festschrift for Morris Halle. Holt, Rinehart & Winston, New York, pp. 377391. Lass, Roger, 1990. How to do things with junk: exaptation in language evolution. Journal of Linguistics 26, 79102. Lass, Roger, 1997. Historical Linguistics and Language Change. Cambridge University Press, Cambridge. Lee, Hyen-hee, 1994. Cwungsey kwuke kwumwun yenkwu (Study on Middle Korean). Sinkwu, Seoul. Lee, Ki Kap, 1987. Miceng uy ssikkuth uli wa keyss uy yeksacek kyochey (The historical replacement of the probability marking uli by keyss). Mal 12, 161197. Lee, Yeong-gyeong, 2005. Siphta kwumwunuy sacek pyenchendy tayhan il kochal (A study on diachronic changes of the siphta construction). Hankwuk mwunhwa 35, 125. r Germanistische Linguistik 32 (2), 152187. Lehmann, Christian, 2004. Theory and method in grammaticalization. Zeitschrift fu Levinson, Stephen, 2000. Presumptive Meanings. MIT Press, Cambridge, MA. volution des formes grammaticales (The evolution of grammatical forms). Scientia 12 (26) reprinted in Meillet, 1958. Meillet, Antoine, 1912. L e ne rale, vol. I. Champion, Paris, pp. 130148. Linguistique historique et linguistique ge Newmeyer, Frederick, 1998. Language Form and Language Function. MIT Press, Cambridge, MA. Norde, Muriel, 2001. Deexion as a counterdirectional factor in grammatical change. Language Sciences 23 (23), 231264. Postal, Paul, 1970. On the surface verb remind. Linguistic Inquiry 1, 37120. Rhee, Seongha, 1996. Semantic of Verbs and Grammaticalization: The Development in Korean from a Cross-Linguistic Perspective. Hankuk publisher, Seoul. Sohn, Semodol, 1995. ko siphtauy uymi cenglip kwaceng (On the semantics of siphta). Kwukehak 26, 147169. Suh, Ceng-Swu, 1996. Kwuke mwunpep (Korean Grammar). Hanyang University Press, Seoul.

1016

M. Kim / Journal of Pragmatics 42 (2010) 10001016

Sweetser, Eve V., 1990. From Etymology to Pragmatics: Metaphorical and Cultural Aspects of Semantic Structure. Cambridge University Press, Cambridge. Thompson, Sandra, Mulac, Anthony, 1991. A quantitative perspective on the grammaticization of epistemic parentheticals in English. In: Traugott, E., Heine, B. (Eds.), Approaches to Grammaticalization, vol. 2. John Benjamins, Amsterdam, pp. 313329. Traugott, Elizabeth C., 2001. Legitimate counterexamples to unidirectionality. Paper presented at Freiburg University, October 17, 2001. Traugott, Elizabeth C., 2002. From etymology to historical pragmatics. In: Minkova, D., Stockwell, R. (Eds.), Studying the History of the English Language: Millennial Perspectives. Mouton de Gruyter, Berlin, pp. 1949. Traugott, Elizabeth C., 2004. Exaptation and grammaticalization. In: Akimoto, M. (Ed.), Linguistic Studies Based on Corpora. Hituzi Syobo Publishing Co, Tokyo, pp. 133156. Traugott, Elizabeth C., Dasher, Richard B., 2002. Regularity in Semantic Change. Cambridge University Press, Cambridge. Vincent, Nigel, 1995. Exaptation and grammaticalization. In: Andersen, H. (Ed.), Historical Linguistics 1993: Selected Papers from the 11th International Conference on Historical Linguistics. John Benjamins, Amsterdam, pp. 433448. Minju Kim is an Associate Professor at Claremont McKenna College. Her research interests include historical linguistics, grammaticalization, discourse analysis, gender and language, and corpus linguistics. Her recent publications have appeared in Studies in Language, Korean Studies, and Korean Linguistics. She received her Ph.D. from University of California, Los Angeles.

Vous aimerez peut-être aussi