On the ‘subject’ honorific -si- in Korean

. Korean is one of several languages that has a system of honorification , whereby a speaker can express (their desired interpretation of) their relative social standing with respect to others via morphosyntactic marking. Alongside the question of what honorification means (what it contributes to the communicative content of an utterance), there is another, less well-studied question: how is the target of honorifi-cation identified? Often this is taken to be a syntactic question, where honorification is treated as a kind of agreement. We present several arguments as to why this is not the correct approach. Instead, we advocate for a semantic/pragmatic solution. We take as our case study one marker of honorification in Korean, the morpheme -si- , which is often called a ‘subject honorific’ because it supposedly targets/honors the subject of the verb it appears on. It is well known, however, that this is an inadequate characterisation, and here we present further evidence that this cannot be the correct description of its role. Instead, we argue that the target of -si- is the human referent which is ‘closest’ to the subject in terms of the pragmatic relation of PROXIMITY , thus accounting for both the canonical uses and other potentially puzzling uses.

1. Introduction. Korean is one of several languages that makes use of a system of honorification, a means of morphosyntactically indicating the relative social standing of discourse participants with respect to each other or to third parties. An interesting difference between Korean and other honorific languages, with the exception of Japanese, is that it relies extensively on verbal morphology in addition to lexical substitutions, whereas other honorific languages rely entirely on the latter (Brown 2015:303). In this paper, we focus on one such verbal marker in particular: the suffix -si-(for a more general overview of the Korean honorification system, see Cho 2022:111ff.). This suffix has traditionally been described as a 'subject honorific', which honors the subject of the verb it attaches to, as can be seen in the following example (taken from Choe 2004:546): 1 (1) Kim Kim sensayng-nim-i teacher-HON-SUBJ o-si-ess-ta. come-SI-PAST-DECL.PLAIN 'Teacher Kim came.' * We would like to thank the audiences at LSA 2023, LAGB 2017, and SE-LFG 23 for their helpful contributions. We would also like to thank our native speaker informants for their time and help. This work was completed partly while one author (Findlay) was in receipt of funding from Research Council of Norway grant number 300495, "Universal Natural Language Understanding", which is gratefully acknowledged. Authors: Yoolim Kim, Wellesley College (ykim6@wellesley.edu) & Jamie Y. Findlay, University of Oslo (jamie.findlay@iln.uio.no). 1 In examples, we use the Yale romanization of Korean, and follow the Leipzig glossing conventions with the following additions: DECL.PLAIN: declarative, of a verb used in the non-honorific, 'plain' hayla-chey speech style; DECL.POL: declarative, of a verb used in the honorific, 'polite' hayyo-chey speech style; DECL.DEF: declarative, of a verb used in the honorific, 'deferential' hapsyo-chey speech style; INT.PLAIN: interrogative, of a verb used in the non-honorific, 'plain' hayla-chey speech style; HON: (nominal) honorific marker, SI: the morpheme -si-, for which we provide an analysis in this paper.
In this case, the subject, Kim sensayngnimi 'teacher Kim', is honored, both by the suffix -nim attaching directly to it, and by the verbal suffix -si-. 2 What exactly does it mean for someone to be 'honored' in this way? We call this the denotational question of honorification; put another way, it asks: what is the semantic contribution of honorification? This has been well studied, with the consensus being that honorification is a type of multidimensional, expressive meaning (Potts & Kawahara 2004; McCready 2019) -specifically, one which expresses a kind of social meaning that denotes a register, cashed out formally as an interval (McCready 2019:ch. 3). That is, when a speaker utters (2), a simple declarative statement with -si-(here realised as the allomorphic variant -usi-), they make two assertions, shown in (2a)  More particularly, the assertion in (2b) means something like "the psychological and/or social distance between me (the speaker) and Sam, and/or level of formality of the speech context, is such that the use of this honorific marker is appropriate" (the relevance of these dimensions to politeness is argued for by Iwasaki & Ingkaphirom Horie 1995;see McCready 2019:28ff. for detailed discussion in the context of honorification). That the expressive part of the meaning shown in (2b) is not part of the at-issue, asserted content 3 can be seen from the fact that it cannot be targeted by direct denial. If one were to reply to an utterance of (2) by saying Aniya thullyesse 'No, that's not true', this could only mean 'No, Sam is not laughing', targetting (2a), and not 'No, you do not honor Sam ', targetting (2b). This analysis of the meaning of honorifics as expressives situates the phenomenon on a by-now wellestablished and burgeoning theoretical foundation, and we will have nothing else to say about the denotational question here.
The second analytical challenge presented by honorification has received much less attention, however; this is the identificational question: how is the target of honorification identified? For some honorific markers this is trivial, since they attach morphologically to their targets (such as -nim, seen above). For markers like -si-, however, which appear on predicates rather than arguments, something further must be said. As noted above, -siis traditionally viewed as targetting the subject of the predicate it attaches to. However, it has long been recognised that this characterisation of -siis inaccurate. For instance, sometimes the target is a possessor embedded inside the subject (Yeon & Brown 2011:189): (3) halapeci-nun grandfather-TOP pang-i room-SUBJ khu-si-ta. big-SI-DECL.PLAIN 'Grandfather's room is big.' Sometimes the target is something not even mentioned in the sentence: 4 (4) i this kkochpyeng-i vase-SUBJ khu-si-ta big-SI-DECL.PLAIN 'This vase (belonging to/created by the honored one) is big.' And, more recently, it has been observed that in customer service contexts -siis used to mark respect towards the customer who is being addressed, even when the subject makes no reference to them (Brown 2015:310): (5) khephi coffee nao-s-yess-supnita. come.out-SI-PST-DECL.DEF 'Your coffee is ready.' This latter use has led some to suggest that -siis shifting towards being an addressee honorific (Kim-Renaud 2001:37).
In Section 3, we discuss the proper characterisation of the target of -si-, and suggest that there is still an important link to the syntactic subject, but that it is mediated via a pragmatic relation of PROXIMITY: the target of -siis the closest human referent to the subject, rather than having to be identical to the it. Section 4 then shows that the predictions this theory makes are borne out by the data: when there is a choice of two potential targets, it is the 'closest' which wins out. Section 5 briefly discusses how our proposal allows for observed dialectal variation, and Section 6 concludes. Before we turn to the main argument of the paper, however, Section 2 takes some time to address the common but, in our view, misguided treatment of honorification as a kind of syntactic agreement.
2. Against a syntactic account. In the generative tradition, it has been common to view the presence or absence of -sias reflecting a kind of agreement (see e.g. Sung 1985;Choe 1988;Kang 1988;Yun 1993;Kim 2007;Choi 2010;Kim 2019;i.a.). That is, certain nouns possess a feature [HON +], and when such nouns are the subject of a a verb, that verb must include the suffix -si-. Such an approach has had success in explaining various syntactic patterns (see Yun 1993:21-30 for some examples, and Brown 2015:309ff. for discussion), but we nevertheless feel that any strictly syntactic approach to honorification is, on conceptual grounds, fundamentally misguided. Honorification is not agreement, and should not be analysed as such. We give here five arguments to this effect. Much of this recapitulates points made elsewhere -see especially Kim & Sells (2007:312ff.) and Brown (2015:310ff.) for overviews.
Argument 1: Agreement is lexically controlled and obligatory; honorification is contextually determined and optional. When a gendered noun induces gender agreement on a dependent adjective, it does so because of some intrinsic feature it possesses; in a language like French, every noun intrinisically bears one of two genders (masculine or feminine). Honorification is very different. It is untenable to suppose that all nouns possess a value for features like [HON ±] that inherently requires/forbids the presence of honorific markers. For example, while the syntactic account might suggest that a noun like sengsayng 'teacher' should bear a [HON +] feature, it is unlikely to suggest that cengpisa 'mechanic' should. But given an appropriate context, for instance where I am discussing a mechanic who came out and repaired my car in a snowstorm, thus saving me from being stranded overnight, it would be entirely appropriate to honor them by using -si-: (6) cengpisa-ka mechanic-SUBJ nemu so chincelha-s-yess-ta. kindness.be-SI-PST-DECL.PLAIN 'That mechanic was so kind!' And the same goes the other way: even if sensayng(nim) is thought to bear the relevant honorificinducing feature in the lexicon, it would be quite natural for an angry student fed up at having received a detention to (rudely) utter the following (see also Argument 3 below on rudeness): 5 In sum, the presence or absence of -siis not determined by lexical properties of the subject; rather, it is a contextually-dependent choice, determined in part by the speaker's attitude towards the subject. This is quite unlike agreement, which is driven by lexical, not cognitive, properties, and is not optional.
Argument 2: Agreement is all-or-nothing; honorification is cumulative. Agreement is discrete: two things either agree or they do not; by contrast, honorification is a matter of degree: more markers of honorification result in a greater degree of respect being ascribed. For example, an Italian verb and its subject cannot agree more or less -they simply agree or fail to agree (the phenomenon is quantal). On the other hand, the sentences in (8) are all identical in their at-issue content, but differ in the level of respect ascribed to the subject -each contains more markers of honorification than the one before, and is concomitantly more respectful (examples from Kim & Sells 2007:315f. By contrast, failing to use an honorific when one is expected, or using one where it is not, is not ungrammatical, but rather results in various effects such as rudeness or obsequiousness. But being rude or obsequious are communicative acts that form part of our linguistic competence, and must therefore be expressible by the grammar. That is, to put it in formal language theory terms, (9b) should not be in the stringset of French (our grammar should not generate it), but (7) should be in the stringset of Korean; rude sentences are part of a language as much as polite ones are.
Argument 4: Agreement does not contribute any meaning; honorification does. When two words agree, this fact in itself does not add any semantic content. That is, if one were to attempt to recover a meaning for an ungrammatical failure of agreement such as (9b), it would be identical to the meaning of the correctly agreeing (9a). On the other hand, (10a) and (10b)  This is related to Argument 3: since the difference between (10a) and (10b) is a meaningful one, the two sentences can have differing communicative effects, relating to social properties like appropriateness/politeness; but since there is no difference in meaning between (9a) and (9b), they cannot.
Argument 5: Agreement is morphosyntactic, but honorification need not target lexically present material. As we saw in (4), the target of honorification can be something not mentioned in the sentence at all. That is, honorification is a relationship between two entities in the world, namely the speaker and the person being honoured. Agreement, on the other hand, is a morphosyntactic relationship between two words. 6 This is a straightforward consequence of the fact that honorification is semantic/pragmatic, not syntactic, in nature.
3. Identifying the target of -si-. Given these arguments, we believe it is clear that honorification as a whole cannot be given a satisfactory analysis as a purely syntactic phenomenon. But does this mean there is no role for syntax at all? The traditional view after all is that the target of -siis identified syntactically: specifically, it is the subject of the predicate which -siattaches to. However, as we noted above, even traditional grammars accept that this cannot be the whole story. In sentences like (11), for example, it is the (denotation of the) possessor of the subject ('Grandmother'), not the (denotation of the) subject itself ('Grandmother's arm') which is the target.
(11) halmeni-kkeyse grandmother-HON.SUBJ pal-i arm-SUBJ apu-si-ta. hurt-SI-DECL.PLAIN 'Grandmother's arm hurts.' The aim of this section is to argue that syntactic analysis is the beginning rather than the end of the process of identifying the target of -si-: there is a further, pragmatic relation between the subject and the target of honorification, namely that of PROXIMITY. Before we explain this in more detail, however, we consider a similar proposal by Kim & Sells (2007). Kim & Sells (2007:310) suggest that the target of -siis not the subject simpliciter, but rather the "maximal human referent" of the subject. What might this mean? On the most obvious interpretation, it would refer to a human referent of which the subject is a part. This would account for (11), since the arm is in a part-whole relationship with Grandmother, the intended target. At the same time, the canonical uses, where the subject is the target, are also explained, since in mereology the parthood relation is usually taken to be reflexive (see e.g. Champollion & Krifka 2016). 7 However, there are also cases where the relationship between subject and target is not mereological, but rather one of alienable possession: (12) halmeni-uy grandmother-POSS chascan-i cup-SUBJ alumda-usi-ta. be.lovely-SI-DECL.PLAIN 'Grandmother's cup is lovely.' As we saw above, this is even possible when the target is not overtly expressed: (13) i this chascan-i cup-SUBJ alumda-usi-ta. be.lovely-SI-DECL.PLAIN 'This cup (belonging to/created by the honored one) is lovely.' were addressees, but that is not the case, as (4) illustrates. It would clearly be unpalatable to extend Kim's (2019) analysis so that all contextually-salient discourse referents are represented in the syntactic tree, but this is what would be required if honorification were truly agreement. 7 We are unclear what is gained by using the adjective 'maximal' in Kim & Sells's definition: we clearly should not project from the human referent connected to the subject up to the maximal mereological sum containing it as a part, or else the target of -siwould always trivially resolve to the sum containing every human in the domain of discourse. If anything, therefore, it might seem more appropriate to talk about the minimal human referent instead. Perhaps the term 'maximal' merely serves as a reminder that one has to potentially look for something 'bigger' than the subject alone.
It does not seem that either Grandmother or whoever owns/created the cup should be considered the maximal human referent of the cup in (12)  This is also a case of alienable possession, and so the subject and the target are not straightforwardly in a part-whole relationship. If scenarios like this are also supposed to fall under the rubric of 'maximal human referent', then the relevant relation must include possession.
There is one class of data which is highly problematic for the 'maximal human referent' approach, however. That is the recently observed phenomenon whereby -siappears to be used to honor the addressee, rather than the subject, as we saw in Section 1. As noted there, this is particularly common in customer service scenarios, illustrated in (15) Here, the subject is i osun 'this article of clothing', which is only connected to the target (the customer) insofar as it might, if it were in stock, have gone on to be purchased by him/her. It is certainly not in a part-whole relation with the target, even on a relaxed interpretation of mereology.
So what is the relevant relation between the subject and the target of -si-? We propose that it is one of proximity: the target of -siis the closest human referent to the subject, in a way to be clarified below. Compositionally, the semantic contribution of -siis as given in (16), where the predicate honored is an abbreviation for the expressive, register-based meaning mentioned in Section 1, but which is not our focus here.
That is, -siis a function which takes the meaning of the subject, x, and asserts that there is some honored human, y, such that no other human is closer to x than y. PROXIMITY(x, y) is a function that returns a value for the 'closeness' of its two arguments (the order of the arguments therefore does not matter) -for the sake of concreteness, we can take this to be a real number between 0 and 1. We intend this semantic/pragmatic relation of proximity to be interpreted fairly broadly; we turn now to what that means. Maximal proximity is identity: it is not possible to be closer to something than being coextensional with it. This gives us the fact that, if the subject is human, then it is itself the target.
We also view possession as a kind of proximity (cf. Barker 1995:46ff.), and assume that inalienable possession is a 'closer' kind of relation than alienable possession. Finally, we suggest that the customer service examples like (15) are instances of a further kind of possession: potential possession. That is, we disagree with the claim that the target in these examples is the addressee per se; instead, we suggest that it is someone who is likely to/intends to come into possession of the subject. This is why the prototypical examples of this usage come from customer service encounters, where a buyer wishes to purchase (i.e. come to possess) something. At the same time, however, such scenarios often mean that the potential possessor and the addressee are one and the same person, making it hard to differentiate the two analyses. Constructed scenarios, however, can tease the two apart. In example (17), we imagine a scenario in which A's father's car has broken down, and A has gone to ask B for help. B has a car to offer A's father, and in their answer, uses -si-: Clearly, the subject itself, i chanun 'this car', is not a suitable target. But, contrary to what we would expect if -sialso has an addressee honorific use, the target is not A either. In fact, the target of -siin this example is A's father, the one who will come to possess the car.
The example in (18) pwutan-i-si-pnita. charge-be-SI-DECL.DEF 'The expense for exchange for a different color will be charged to the (honored) buyer.' Here, the target is the buyer, who will have to pay extra for exchanging the color of whatever they are purchasing. The buyer is present in the sentence (kwumaycanim), but it is not the subject (which is sayksang kyohwan piyongun 'color exchange cost'). Kim & Sells do not provide any context for this example, but we can certainly imagine it occuring in a situation where the buyer is not present, and so is not the addressee. What licenses the buyer as the target of -siin that case is simply that they are the one who will incur the cost, i.e. come to possess it.
In addition to the conceptual or social proximity represented by various kinds of possession, we might wonder whether the notion of proximity relevant for identifying the target of -sican also be straightforwardly spatial. On the one hand, it is not possible to honor someone who simply happens to be nearby to the subject, so physical proximity alone is not sufficient: Here the subject is kunmwu kanung ciyekun 'possible work area', which is not human and so not a viable target. The actual target is not a possessor, but rather someone who will work in the area, i.e. be in close physical proximity to it (being inside the area is presumably at basically the highest degree of physical proximity short of identity or parthood).
What is the difference between (20) and (19), then? The facts are not yet clear to us, but one potentially important difference is that there is a lexical-conceptual connection in (20) between the target and the subject: the person honored is present at the work site in order to work. That is, perhaps the proximity of target and subject must also be motivated by something in the subject's meaning, in the broader sense of 'meaning' covered by the notion of qualia in generative lexicon theory (Pustejovsky 1995). It's not just any spatially proximal human that can be the target, but only one which participates in some (potentially 'hidden', i.e. syntactically unexpressed) event that the subject makes available. We leave further analysis of this restriction to future work, and focus here on cases where conceptual rather than purely physical proximity is at stake.

4.
A scale of conceptual proximity. Identity plus the three kinds of possession discussed above can be arranged in a hierarchy of descending proximity, shown in Figure 1. As stated above, identity is maximal proximity. Being part of something, or otherwise inalienably connected to it, is not quite as close as being identical to something, but still closer than being separable, as in alienable possession. Lastly, mere potential possession ranks lower than actual possession. Such a scale makes predictions: when possible targets are in competition, the target which sits highest on this scale of proximity should win out. 8 We turn now to several examples which show that the data bear this prediction out. (21), there are two human referents mentioned inside the subject, emeniuy 'mother' and salincaka 'murderer':

IDENTITY > INALIENABLE POSSESSION. In a sentence like
(21) (#) emeni-uy mother-POSS salinca-ka murderer-SUBJ canin-ha-s-yess-ta. cruel-be-SI-PST-DECL.PLAIN 'Mother's murderer was cruel.' Taken in the abstract, the far more plausible target of honorification here is the mother: parents are prototypical targets of honorification, and murderers very much not. However, since identity outranks inalienable possession in terms of proximity, the actual, linguistically-determined target of -siis the murderer, making (21) a decidedly odd sentence to utter.
This shows that the identificational problem cannot be reduced to a general pragmatic strategy for finding the most 'plausible' target, or something similar. There is something more mechanical at work here, that can lead to clashes between the output of the linguistic system and what our real-world knowledge leads us to expect.

INALIENABLE > ALIENABLE POSSESSION.
To illustrate the next step in the scale, we evoke a slightly exotic context. Let us say that father was a lover of science, and left his body to medical research on his death. He also had a very distinctive tattoo on his finger. In my human anatomy class, where each of us has a hand to dissect, and the professor is demonstrating on a hand of his own, I notice that very same tattoo, and say the following: There are two competing human targets here: the inalienable and alienable possessors of the hand, viz. my father and the professor. In this context, the target of -siis my father, the inalienable possessor, in keeping with the scale of proximity. This is in spite of the fact that kyoswunimuy 'professor' appears as part of the subject, while apeciey 'father' does not, further illustrating the inadequacy of a wholly syntactic account, and reinforcing the importance of the pragmatic relation of the target to the subject. 4.3. ALIENABLE POSSESSION > POTENTIAL POSSESSION. To consider the relative ranking between alienable and potential possession, we use a slightly modified version of example (17) (17) it did not. There are therefore two potentially eligible human targets in play: the potential possessor of the car, A, and the alienable possessor, B's father. As the scale would predict, the target here is B's father, not A. 4.4. SUMMARY. This section has illustrated that the process which determines the target of -siis sensitive to the conceptual proximity of that target to the subject. The fact that proximity is a scale means that we make predictions about what happens when there are multiple potential targets, unlike other approaches which assume that the subject can be straightfrowardly resolved to a single human target. In such a case, it is the 'closest' potential target to the subject which is chosen, even when that is otherwise implausible, e.g. based on real-world knowledge.

Dialectal variation.
There is variation among Korean dialects as to how 'distant' a permissible target can be from the subject. While all varieties allow identity with or inalienable possession of the subject, for many speakers, alienable possession and potential possession are not possible. This is illustrated in Figure 2. The nature of this variation lends further weight to our scale of conceptual proximity, since the division between more conservative and more liberal dialects aligns with continuous segments of it: there is a cut-off point in terms of proximity. We do not, for instance, observe dialects which permit some discontinuous portions of the scale, e.g. allowing identity and alienable possession as possible relations, but not inalienable or potential possession.
Formally, we can say that the point of differentiation between these two groups of dialects is in the definition of the PROXIMITY predicate: in the liberal dialects, it is defined over all kinds of possession, whereas in the conservative dialects, it is undefined over alienable and potential possession (or alternatively returns the minimum value, viz. 0, if PROXIMITY is taken to return a real number between 0 and 1). In these dialects, the only interpretation for a sentence like (12), repeated below as (24), is the anthropomorphic one in which the subject, halmeniuy chascani, i.e. the cup itself, is coerced into a human interpretation.
(24) halmeni-uy grandmother-POSS chascan-i cup-SUBJ alumda-usi-ta. be.lovely-SI-DECL.PLAIN 'Grandmother's cup is lovely.' 6. Conclusion. This paper has made two main contributions. Firstly, we presented several arguments against the the common syntactic account of honorification, which treats it as a kind of agreement. Secondly, we argued that the target of -siis identified with reference to the pragmatic relation of PROXIMITY: the target of -siis the human being which is 'closest' to the subject. This notion of closeness incorporates conceptual proximity, which includes different kinds of possession. Our proposal stands apart in making testable predictions about what happens when there are multiple potential targets, and these predictions are borne out by the data. In future work we plan to explore other factors which interact with proximity (e.g. the animacy of the subject), and conduct survey work to test further variation between speakers.