Postsyntactic Lowering and linear relations in Dagur noun phrases

In this paper, I investigate the unusual order between case and possessive morphemes in the endangered Mongolic language Dagur. It is observed in many languages that the case morpheme follows possessive markers, but Dagur uniformly exhibits the morpheme order where possessive follows case in its nominal domain. Based on novel data from fieldwork, I propose that such order is due to postsyntactic Lowering, in which the head of K(ase)P lowers to Poss(essive) head. The evidence for the Lowering analysis comes from suspended affixation in this language. In particular, suspended affixation involving case and poss morphemes displays unusual patterns, compared to the suspension of other types of morphemes. Further investigation reveals that Dagur suspended affixation is best analyzed as a base-generated structure, instead of morpheme ellipsis. Given these, the Lowering analysis correctly derives the poss-final order and successfully accounts for the unusual patterns in suspended affixation. In addition, I examine surface morphophonological differences between the case and the poss morphemes, and show that they can be systematically encoded in the current theory.


Introduction
In languages with rich inflection, overt case morphology is often observed at the periphery of the nominal domain. In terms of hierarchical structures, case has sometimes been taken to be the maximal extension of the nominal projection (e.g., Siegel 1974;Lamontagne & Travis 1987;Bittner & Hale 1996). In Dagur, an endangered Mongolic language, the case suffix uniformly appears to be closer to the stem than the possessive suffix. This is unexpected if we assume case to be the highest head of the extended nominal projection, and that linear order mirrors the order of syntactic heads and projections (e.g., Baker 1985). The linear order which directly maps onto levels of hierarchical projections is not the only linearization possibility. Embick (2015), for instance, suggests that whether or not the morpheme order mirrors the levels of syntactic projection depends on how the relevant morphemes are linearized. Specifically, the output of syntactic derivation may also be subjected to a series of PF operations that further manipulate the structure, conditioning the final linearized output. The current analysis of the Dagur morpheme order provides a case study in which the surface linear order does not directly mirror the underlying hierarchy of projections, but is related to it in a systematic way. The proposal, couched in the theory of Distributed Morphology, derives the Dagur suffix order via a series of operations which take place postsyntactically. In the first part of my analysis, I examine the syntactic structure of the Dagur nominal domain, and propose a postsyntactic Lowering operation which lowers K onto Poss, in light of the data from suspended affixation. In particular, I show that suspended affixation involving case and poss morphemes displays unexpected patterns, in that they are not predicted by the right edge constraint on suspended affixation. Assuming that K is the highest head in the extended nominal projection, I demonstrate that suspended affixation is base-generated from a structure in which two smaller constituents are coordinated under a single morpheme, with further inflections taking place on that morpheme. Lowering of K to Poss applies postsyntactically to the base-generated structure, successfully capturing all the possible and impossible surface forms of suspended affixation. In the second part of my analysis, I examine further morphophonological differences between case and poss morphemes. Specifically, the case morpheme appears to be morphophonologically closer to the stem than the poss morpheme. I present one specific proposal which utilizes morphological word formation based on the given syntactic structure, in combination with linear free rebracketing. The rebracketing operation is also shown to be systematically connected with the Lowering operation which applies before linearization. Finally, I will discuss the implications of my analysis for the general architecture of postsyntactic operations.
The rest of the paper is organized as follows. In Section 2, I introduce the main empirical phenomenon. In Section 3, I present detailed fieldwork data on suspended affixation and argue that it should be analyzed as a base-generated structure rather than ellipsis. In Section 4, I outline my assumptions for the syntactic structure of Dagur nominal domain. Given this structure, I present the main proposal in which the stem-case-poss order is derived by K undergoing Lowering to Poss after syntactic derivations. In Section 5, I examine further morphophonological properties of the case and the poss morphemes, and suggest several morphological processes to derive these properties. Section 6 concludes the paper.

The poss-final order in Dagur
In Dagur possessive constructions, the suffix following the possessum agrees in person and in number with the possessor. The prenominal possessor is often optional, but the possessive suffix is obligatory.
(1) a. When the possessive construction is marked for case, the case suffix precedes the poss suffix. The reversed order is ungrammatical.
Within the Mongolic family, the surface order in which poss follows case has also been observed in other languages. As shown in (3), the accusative case morpheme seems to precede the first person possessive morpheme in Khalkha Mongolian, a language closely related to Dagur. (3)

Khalkha Mongolian
Baatar nom -ig miny unsh -san  (4b), it surfaces as a phonologically reduced form miny (often regarded as a particle). Crucially, the postposed miny in (4b) still remains a full pronoun, because the prenominal minii and the postnominal miny can never co-occur (4c). This suggests that miny, as a pronoun, cannot be locally bound, and (4c) is ungrammatical due to Condition B violation.
Since the possessive pronoun in Khalkha can surface postnominally, the fact that in (3) 1sg. poss linearly follows case is not surprising. In standard Khalkha orthography, the postposed pronominal possessor such as miny in (3) is also written as a separate word. In contrast, evidence from binding shows that poss markers in Dagur exemplify a different case, in that the poss morpheme signals agreement with the possessor. In light of this, part of this paper's empirical task is to determine a) the morphological status of poss with regard to the rest of the nominal domain, b) the syntactic status of poss within the extended nominal projection, and c) the similarities and differences between poss and case in these respects. Answering these questions would ultimately help us determine the underlying mechanism that is responsible for the morpheme order observed on the surface. In the remainder of this paper, I divide this puzzle in Dagur into two parts: the hierarchical syntactic structure and the post-linearization morphological processes. I will begin with investigating the underlying structure of Dagur nominal domain and the morphosyntactic properties of the nominals with the case-poss suffix order. Through a detailed investigation into the phenomenon of suspended affixation, I will show that a) suspended affixation in Dagur is not the result of ellipsis. Instead, it is base-generated from a structure where two smaller constituents are coordinated under a single morpheme; and b) noun phrases followed by case and poss suffixes display unusual morpheme suspension patterns, which can be adequately accounted for under the view that case, occupying K(ase), undergoes postsyntactic Lowering to the Poss head, which hosts the poss morpheme. Then, I will turn to surface morphophonological properties of the case and the poss morphemes, and discuss possible ways to derive them under the current theory.

Background
In this section I introduce the phenomenon of suspended affixation in Dagur, and present novel data which indicate that it is base-generated from a structure where the suspended affixes attach to the entire coordinate structure. Suspended affixation is first defined by Lewis (1967) as the construction in which one grammatical ending serves two or more parallel words. In suspended affixation constructions, the final conjunct bears one or more word-final suffixes, and the non-final conjunct lack those suffixes (Kornfilt 2012). In Dagur, nominal phrases can be coordinated with or without an overt coordinator, and both types allow suspended affixation. 1 (6) Seb boloor šeb j -sul ir -sen.
worker, farmer, teacher, student -pl come -pst 'Workers, farmers, teachers, and students came.' * 'A worker, a farmer, a teacher, and students came.' In (6), two noun phrases are conjoined by the coordinator boloor, with pl suffix -sul following only the final conjunct. However, both conjuncts in (6) have plural interpretation. In (7), there is no overt coordinator, but the same pattern is observed.
A typical Dagur nominal stem may be followed by three types of markers: pl, case, and poss, and any one of these three can undergo suspended affixation. Suspension of pl suffix has been illustrated with (6)-(7). The following two examples show suspension of case (8) and poss (9) morphemes respectively.
(8) Engkebatu (1988) 1 In most cases, coordinate structures with overt coordinators and those without overt coordinators do not have significant interpretative differences.

2
Dagur lacks a determiner system like the one in English. In Dagur bare singulars are number neutral. In order to express 'a teacher, and students came', the following structure can be used: When all three suffixes are present in a nominal phrase, the linear order is rigidly stem-plcase-poss, illustrated with (10). Suspended affixation involving more than one suffix will be discussed in detail in the next section.

Further data on suspended affixation in Dagur
Having introduced the basic structural properties of suspended affixation in this language, I present further morpheme suspension facts in this section. One of the key observations is that most instances of suspended affixation in Dagur obey the right edge constraint observed crosslinguistically. However, suspension involving case and poss morphemes displays unexpected patterns, suggesting further structural complications at play. The set of data presented here contributes to a larger empirical picture on the cross-linguistic variation of the morpheme suspension effects, in which the patterns differ according to language and morpheme-specific factors (e.g., Kornfilt 1996;2012;Despić 2017;Yoon 2017;Guseva & Weisser 2018;Erschler 2018).
I begin with the illustration of the coordinations of words with only one suffix. As presented in the previous section, if each conjunct is only followed by one suffix (pl, case, or poss), then suspended affixation can freely apply. (16) *Bi seb -tii boloor šeb j -sul -tii usɣulj -sen -bi.
I teacher -comit conj1 student -pl -comit talk -pst -1sg Int. 'I talked with teachers and students.' When each conjunct is followed by poss and pl, poss follows pl suffix, and the morpheme suspension pattern is the same as the previous illustration involving case and pl -poss suffix can be independently suspended (17). Alternatively, both pl and poss can be suspended (18 As shown in Table 1, if each conjunct is followed by only one suffix, then suspended affixation can freely apply. If each conjunct is followed by two suffixes, then the suspended suffixes must be the rightmost in its conjunct. So far, this pattern appears to be an instance of a more general empirical observation termed the right edge condition (e.g., Guseva & Weisser 2018), widely observed in languages with suspended affixations. (20) The right edge condition: The elements omitted due to suspended affixation must be at the right edge of the non-final conjuncts.
However, when the coordination conjoins words marked by both case and poss suffixes, suspended affixation displays unexpected patterns given the observations in Table 1 and the generalization in (20). In (21), the rightmost suffix -maan j is suspended, which conforms to the right edge condition. However, this sentence is ungrammatical.
Mergen house -dat conj1 school -dat -1pl.poss go -pst Int. 'Mergen went to our home and our school.' If only case is suspended as in (22), the sentence is also ungrammatical. However, case and poss can be suspended together (23).  A complete list of suspended affixation patterns in Dagur nominal domain is presented in Table 2. Table 2, suspended affixation involving both case and poss morphemes (rows j., k., l. of Table 2) displays unexpected patterns in the sense that it is not fully captured by the right edge constraint. In particular, while poss is located at the right edge of each conjunct, suspending poss independently is unacceptable.

Background
Depending on the language, the structure of suspended affixation can be analyzed in various ways. One of the analyses proposes it to be ellipsis (e.g., Erschler 2018; Guseva & Weisser 2018), in which two fully inflected forms are coordinated, with the suspended affixes elided at PF: Another approach (e.g., Kornfilt 1996;Good &Yu 2005;Kabak 2007) analyzes suspended affixation as a coordinate structure conjoining two smaller constituents, with further inflection taking place on that coordination. This is in a similar vein with Kornfilt's (2012) right node raising analysis of Turkish suspended affixation. I refer to this line of analysis as the basegeneration account.
(29) base-generation analysis We could first consider how the analysis that utilizes morpheme ellipsis fits into the Dagur data at hand. This is evaluated through a brief empirical comparison between Dagur and Mari (Finno-Ugric), reported in Guseva & Weisser (2018) (henceforth G&W (2018)). I focus on one specific fact in Mari, where "local" case such as inessive case precedes the possessive suffix, exhibiting the case-poss linear order which is also attested in Dagur:  stem-case-poss & stem-case-poss * (case suffix refers to a subset of cases which G&W refer to as K1 or local case)

Conjunction Example#
In G&W's account, the case-poss order is the consequence of the interaction between morpheme ellipsis and a linear metathesis operation (D-metathesis in G&W:1105). G&W assume a uniform DP-internal syntax and a mapping algorithm such as the Mirror Principle. Assuming the levels of projection is underlyingly (32) in Mari, the direct output of linearization is stem-pl-poss (D)case (K). (32) The underlying structure of Mari nominals G&W argue that the facts in (31) are due to the original right edge suffix (i.e., -case) undergoing ellipsis after linearization, but crucially before the metathesis operation changes the order from poss-case to case-poss. Thus case -the right edge suffix before metathesis -can be elided, and the right edge condition is satisfied at this intermediate stage (i.e., at (33b)). 4 The key point of this component of their analysis is that morphological operations (ellipsis and metathesis) apply to the linearized structure, deriving the suspension patterns that appear to violate the right edge condition on the surface form. However, such mechanism does not easily extend to Dagur. Regarding the stem-case-poss constructions, while in Mari suspending the non-final case suffix is grammatical (34a), it is ungrammatical in Dagur. An analysis that utilizes ellipsis and metathesis on linearized structures would not be able to rule out both surface forms (34a-b) at the same time.
coordination with suspended affixation Mari Dagur Further investigation into the syntax prior to linearization reveals that Dagur suspended affixation involves a syntactic structure that is different from Mari. I suggest that it is this syntactic structure in combination with other postsyntactic operations that generate the Dagur surface pattern in (34). 3 Here "stem" abstractly represents the [[Root n]-pl] complex, or [Root n]. As the presence of plural does not affect the point being made here, it is omitted for expository purpose.

4
It should be emphasized that this discussion is a significantly oversimplified illustration of one aspect of G&W's theory. The morphological processes in Mari are complex and require the combination of several postsyntactic operations. For example, there is a separate Lowering operation that derives the order between poss and pl. pl also intervenes between case and poss which leads to a specific formulation of metathesis operation called D-Metathesis (G&W: (41)). However, these do not affect the general point being made here.

The base-generated structure of Dagur suspended affixation
In contrast with recent analysis of languages with morpheme ellipsis (e.g., Erschler 2018; Guseva & Weisser 2018), independent evidence from different types of coordinators reveals that suspended affixation constructions in Dagur is best analyzed as base-generated in the sense of (29), instead of ellipsis (28). In Dagur, different types of coordinate structures require different coordinators. Specifically, coordinated nominals in argument positions are conjoined by boloor, whereas the coordinator beitleen is used when the conjoined elements of nominal, adjectival, and verbal categories are in non-argument positions. Throughout the paper, I gloss boloor as conj1, and beitleen conj2 in order to differentiate between the two. I illustrate the distribution of these two coordinators in detail below, focusing on the coordinate structures of nominals and adjectives. The properties of these coordinators ultimately offers us a testing ground for the ellipsis versus the base-generation analysis of suspended affixation in this language.
First, as previously mentioned, boloor (conj1) is used to conjoin nominals in argument positions. This is shown in (35)-(37), with coordinated bare nouns, plural nouns, and possessed nouns in argument positions. I apple -acceat -pst conj2 hackberry -acc also eat -pst -1sg 'I ate apple and (also) ate hackberry.' Two points regarding the usage of these coordinators should be highlighted. First, while in both (35) and (38a) there appears to be two noun phrases that are coordinated, the structure of (38a) is clearly distinct from that of (35). Specifically, (38a) is a predicative sentence in which the coordinated noun phrases seb 'teacher' and tačkui daa 'principal' are non-referential NPs that function predicatively. The sentence ascribes the individual Sečin the property of being a teacher and a principal. Thus, the coordinator beitleen (conj2) must be used. (38a) could be contrasted with (39), in which the coordinated noun phrases are located in the object position of the main verb uǰsen 'saw'. Since the noun phrases are arguments of the verb, the coordinator boloor (conj1) must be used. The coordinated noun phrases in (39) refer to two individuals, in contrast to (38a). 5 The contrast among (38a), (39), and (40) shows that the choice of coordinators in Dagur is sensitive to whether the coordinated nominals occur in argument positions or not. Specifically, boloor (conj1) is used to conjoin nominals in argument positions, whereas beitleen (conj2) is used when the coordinated elements appear in non-argument positions.
The second point regarding the usage of the coordinators concerns the interpretation of coordinated adjectives. Generally speaking, adjectives can appear in two main types of syntactic contexts -as attributive adjectives directly modifying a noun, or as predicates (see Valois (2008) and Hofherr (2010)  this room spacious conj2/*conj1 bright 'This room is spacious and bright.' As indicated by the translation, in (41) the only reading available is one in which both adjectives jointly modify the noun 'apple'. That is, the meaning is necessarily '(an) apple that is big and red', not '(an) apple that is big and (an) apple that is red'. Using the argument coordinator boloor (conj1) in (41) is not acceptable. 7 Given that only beitleen (conj2) is allowed as the overt coordinator in (41), and that the structure only licenses joint reading, where 'big' and 'red' modify the same apple, a plausible structural analysis for beitleen-coordinated prenominal attributive adjectives seems to be that it does not involve underlying coordination of two nominals. Instead, it is two adjectives that are coordinated in constructions such as (41).
With the distribution of different coordinators in mind, we are now ready to test whether Dagur suspended affixation involves morpheme ellipsis, or a base-generated structure without 6 When the 3sg.poss suffix -in j is used without an overt possessor, the NP is ambiguous between a pure definite (without possessive) interpretation and a possessive interpretation. For example, biteɣ -in j (book-3sg. poss) means either 'the book' (referring to a book salient in the context or previously mentioned in the discourse, with no possessive interpretation), or 'his/her book' (implying there is a third person possessor). When there is an overt possessor, the construction is unambiguously possessive. For example, Sečin-in biteɣ-in j (Sečin-gen book-3sg.poss) simply means 'Sečin's book'.

7
A form such as (i) can be used to obtain the reading '(a) big apple and (a) red apple'.
(i) xiɣ pinguee boloor xulaan pinguee big apple conj1 red apple 'big apple and red apple' ellipsis. The crucial data involve suspending the 3sg.poss suffix -in j in a special nominalizing context. In addition to signaling possessive agreement, the suffix -in j can change an adjectival phrase into a noun phrase, with no possessive interpretation. For example, a regular adjective xulaan ('red', as in (43a)) followed by -in j gives rise to the reading of 'the red one' (43b-c), with no possessive interpretation. Without -in j , the adjective cannot function as an argument (43d). This "nominalizing" -in j is in complementary distribution with regular possessive suffixes. As shown in (44), the special -in j morpheme and regular possessive marker cannot be stacked. I take this to indicate that the special -in j and regular possessive morphemes occupy the same head in the nominal structure. Given the above facts, the coordinate structure (45a) in which two "nominalized" adjectives like the one in (43b) are conjoined requires the coordinator boloor (conj1), not beitleen (conj2). The choice of coordinator in (45a) is expected since boloor conjoins two nominals in the argument position, whereas beitleen does not. Surprisingly, if the suffixes on the first conjunct is suspended, the resulting construction in turn requires the coordinator beitleen (conj2) as in (45b), and is ungrammatical with boloor (conj1), shown in (45c). (I) big conj1 red -acc -3sg.poss eat -pst -1sg We know that the choice of coordinators is sensitive towards the argument/non-argument status of the nominals. If suspended affixation were ellipsis, the coordinations in (45a)  would share the same underlying structure which is schematized in (46), and would require the coordinator boloor (conj1) instead of beitleen (conj2). However, the fact turns out to be that once suspended affixation takes place, the coordinator boloor (conj1) becomes ungrammatical (45c), and instead the other coordinator beitleen (conj2) is required (45b). These facts would not be straightforwardly explained if suspended affixation is morpheme ellipsis. In contrast, they are fully expected if the morpheme suspension phenomena observed on the surface structure is in fact base-generated in the syntax. Under this analysis, the structure in (45b) is in fact (47), where two adjectives are conjoined. As a result, only beitleen (conj2) is allowed as the coordinator. Before presenting the main analysis, I first examine some additional characteristics of the Dagur nominal domain that will help us determine its specific structure in the syntax. Bošković (2005;2010a) and research thereafter show that languages without articles differ from languages with articles in systematic ways. Under the view that the lack of definite article suggests the lack of DP, Dagur, lacking an article system, can be regarded as an NP language. 8 If we adopt this view, some of the empirical phenomena we have seen in the previous section naturally follows. First, we have seen that bare nouns directly function as arguments in this language. This contrasts with DP languages such as Romance languages, in which NPs cannot occur as arguments unless they project a category D(eterminer). In Dagur, the bare noun followed by various nominal suffixes may also directly function as arguments.  Engkebatu (1985) …duut terwul -ii eri -sen.
shortcut road -acc search -pst '(She/he) searched for the shortcut.' Second, the location of coordinated APs modifying a noun (e.g., (41), repeated here as (49)) can be naturally related to the internal structure of NP. Bošković (2005) argues that the position of AP in Serbo-Croatian (SC), an NP language, differs from the one in DP languages such as English. He suggests that DP languages exhibit the AP-over-NP pattern, whereas NP languages have the NP-over-AP pattern. According to this view, Dagur is expected to have the NP-over-AP pattern, and the structure in (49) would involve a pair of coordinated APs adjoined to the [ np 8 As an anonymous reviewer points out, many authors have argued that the lack of articles does not always indicate the lack of DP. For arguments made based on evidence from different languages, see Manlove (2015) for West Greenlandic, Lyutikova & Pereltsvaig (2015) for Tatar, Giusti & Iovino (2016) for Latin, Stanković (2017) for Serbo-Croatian, Syed & Simpson (2017) for Bangla, Norris (2018) for Estonian, and Erschler (2019) for Ossetic. However, many of these proposals cannot be easily extended to Dagur, which seems to lack empirical support for positing a DP layer in its nominal domain. Therefore I will assume Bošković's approach for the purpose of this work.
For the general debate on the very existence of DP, see Bruening (2009; and Chomsky (2020) for arguments against the DP, and the replies of Preminger (2020) and Salzmann (2020) to Bruening's arguments.

Gong
Glossa: a journal of general linguistics DOI: 10.5334/gjgl.1397 apple]. Since the structure involves a pair of conjoined APs modifying one single NP, the only possible interpretation is 'a single apple/a single set of apples that is both big and red '. 9 (49) xiɣ beitleen xulaan pinguee big conj2 red apple '(a) big and red apple(s)' Dagur's nominal structure has implications for the location of possessives within the nominal domain as well. It has been argued that in some NP languages such as SC (Zlatić 1997;Bošković 2005;), possessors pattern with APs in many ways and thus should be analyzed as modifiers of NP. However, the possessors in Dagur do not pattern with APs as in SC, suggesting that they should not be simply treated as modifiers of NPs. First, in Dagur noun phrases, the possessor (if available) is always the first element, followed by demonstratives, which is in turn followed by numerals, adjectives, and the head noun. The order between numeral and adjective could be flexible, but neither may precede the demonstrative or the possessor. The demonstrative may not precede the possessor. The head noun usually does not bear any plural suffix if a numeral precedes it. An example is given in (50).

Theoretical assumptions
The current analysis is couched in the framework of Distributed Morphology, along the lines of Embick & Marantz (2008) and Embick (2010;. In particular, the syntax creates complex objects out of different types of morphemes, the Roots and the functional morphemes, which are terminal nodes in the syntactic derivation. The Roots are assumed to be category-neutral, and are categorized in the syntax by category-defining heads such as n, v, a, yielding nouns, verbs, adjectives, and so on. These category-defining heads are regarded as cyclic in the sense of the phase theory, following the view that syntactic derivations operate in terms of cyclic domains (Chomsky 2000;2001). I assume that in Dagur, a noun phrase such as the one in (56) has the levels of projection as depicted in (57) Embick & Noyer (2001). Vocabulary Insertion at PF supplies syntactic terminal nodes with phonological content. For example, the Vocabulary Item in (58) applies to the node #[+pl] for regular nouns such as mori 'horse' and seb 'teacher': Similarly, by Vocabulary Insertion poss morphemes are inserted at the Poss head, and case at K. In the rest of this section, I will demonstrate that the hierarchy in (57b) can be maintained if K is lowered to Poss after syntax in the PF component.

Account for the morpheme suspension patterns
Based on cross-linguistic evidence, research such as Lamontagne & Travis (1987), Bittner & Hale (1996), McFadden (2004, and Levin (2015) takes the highest head in the nominal projection to be K. This is also what the current paper assumes in the underlying structure (57b). However, directly linearizing this structure will give rise to the order in which case follows poss, contrary to the surface case-poss order observed in Dagur nominals. I suggest that Lowering, a post-syntactic morphological operation, helps produce the surface linear order in which poss follows case.
(59) Lowering of X 0 to Y 0 ( Embick & Noyer 2001) [ The Lowering operation explains the unexpected suspended affixation patterns in constructions with both case and poss present. In particular, in j., k., and l. of Table 2, repeated here, when the suffix sequence -case-poss is present on the noun, suspending either case or poss is unacceptable. The only grammatical option is one in which case and poss are suspended altogether (the presence or absence of an additional plural morpheme does not affect the judgment of these constructions, thus it is omitted in the representations in Table 3).
In the analysis proposed here, the data which satisfy the right edge condition (i.e., row j) but are nevertheless ungrammatical are, in fact, violations of the underlying levels of projection. In addition, the ungrammatical data which violate the right edge condition (i.e., row k) can be independently explained from the fact that the coordinate structure blocks K-Lowering. The process of K lowering to Poss at PF is illustrated in (60)  First, the surface form [✓ stem-case-poss & stem-case-poss] in row l. of Table 3 is derived by conjoining two stems, with Poss and K directly merging above the coordinate structure. K-Lowering operates on the structure (62a), giving rise to (62b): (62)  Table 3. On the surface, this form is ungrammatical because, as described by the right edge condition, suspending the non-final suffix is not allowed. Under the current analysis, this surface condition can be independently explained on the basis of the syntactic structure. As shown in (64), the coordinate structure in which two PossPs are conjoined below K blocks K-Lowering. The reason is that by the definition of Lowering (59), a head only lowers to the head of its complement. Here the head of K's complement is the coordinator head. Therefore K cannot lower to Poss.

Int. 'At our home and (our) school'
Note that it is not the case that KPs cannot be coordinated in this language. This is an important point to clarify, because the current analysis suggests that (66) is unacceptable because of the constraint on the underlying syntactic structure, which states that KP must dominate PossP. In order to maintain this analysis, it must be independently shown that the language allows [stemcase] & [stem-case], and thus it is not the case that (66) is ungrammatical due to a ban on conjoining two KPs in the language. Coordinating KPs is indeed allowed in Dagur. For example, (69) is grammatical: (69) (Bi) seb -d, guč -d ǰašɣen ši -sen -bi (I) teacher -dat friend -dat letter write -pst -1sg 'I wrote letter to teacher(s) and friend(s).'

Comparison with linear displacement analyses
As demonstrated, the analysis based on Lowering accounts for all the possible and impossible forms of suspended affixation involving case and poss morphemes. At this point, it is worthwhile to compare it with other analytical possibilities. For example, one alternative is to consider linear morpheme displacement such as Local Dislocation (e.g., Embick 2007a;b; Embick & Noyer 2007;2001). Similar to Lowering, Local Dislocation is a postsyntactic operation that is responsible for surface displacement of morphemes that deviates from the order of levels of projections in syntax. While Lowering applies to the hierarchical structure prior to linearization, Local Dislocation takes place after linearization and manipulates stringadjacent elements.
Historically, it seems that the Dagur personal possessive pronouns used to appear in postnominal enclitic positions, which has resulted in the grammaticalization of poss suffixes. 10 Since there are diachronic reasons for thinking that poss suffixes are developed from postnominal enclitics, an analysis in which poss is directly displaced to a position in which it linearly follows K appears to be plausible. Take (70) as an example, suppose that no Lowering operation applies at the syntactic structure and linearization takes place right after syntax. Linearization would produce a statement such as the one in (70b). The notations in (70b-c) follow Embick & Noyer (2001) and Embick (2007a) and use a * b to denote linear adjacency between heads and phrases (stating that a must linearly precede b and be adjacent to b), and a ⊕ b to represent adjacency within complex heads. Crucially, Local Dislocation makes reference to linear precedence and adjacency, which are encoded in the linear statement (70b), not in the hierarchical structure, which is (70a). Therefore Local Dislocation converts (70b) to (70c), giving rise to the surface order where Poss linearly follows K. It appears that the Local Dislocation analysis, without resorting to Lowering, can also derive the grammatical surface form (70). 10 The (genitive) personal pronouns have been used as enclitics in Proto-Mongolic, which has resulted in the grammaticalization of poss suffixes in several modern Mongolic languages (Janhunen 2003). Dagur's poss agreement marking system is an example of such grammaticalization. The high resemblance in the morphophonological shapes of personal pronouns and poss markers also suggest a close connection between the two: gen pronoun poss marker gen pronoun poss marker 1s minii -min j 1pl maanii -maan j 2s šinii Since the poss markers under investigation used to be independent words, it displays a certain degree of phonological independence compared to regular suffixes like case, a point which I will come back to in Section 5. However, such analysis makes incorrect predictions regarding the surface form (71), which has the hierarchical structure (71a) in syntax, where two PossPs are coordinated under a single K head. Linearizing (71a) produces the statement in (71b). In (71b), the boldfaced Poss is the right-peripheral element of the constituent that is immediately adjacent to K. Since Local Dislocation makes reference to linear adjacency, not hierarchical structure, it has the capacity to convert (71b) to (71c), giving rise to the surface string [ nP-Poss & nP-K-Poss ] (i.e., [stem-poss & stem-case-poss]). However, this form is ungrammatical in Dagur. Note that without additional stipulation, the application of Local Dislocation to (71b) is legitimate since all the linear precedence or adjacency (*) relations in (71b) are either preserved or properly converted in accordance with the Local Dislocation rule. Prior to Local Dislocation (71b), the second nP must be linearly followed by Poss. After Local Dislocation (71c), such relationship is maintained -this nP is still linearly followed by Poss, which is now internally complex. In other words, without additional stipulation, Local Dislocation would apply to the structure in (71), and generates an ungrammatical form. In contrast, the Lowering analysis correctly rules out (71), on the basis that the coordinate structure blocks the higher K head from lowering to Poss.
(71) *[stem-case-poss & stem-case-poss] ( The discussion above potentially provides another case study to the literature on the distinction between hierarchical and linear morphological processes and the ordering among postsyntactic operations. The reason why Local Dislocation and Lowering make different predictions with regard to (71) ( Table 3, row k.) is that Lowering, applying before linearization, is sensitive to the syntactic structure, particularly the levels of projections and syntactic headedness. In contrast, Local Dislocation, taking place after linearization, makes reference to the linear order and is therefore sensitive only to linear precedence and adjacency. Since Local Dislocation must always be local, K and Poss located on the right periphery of (71b) would satisfy such requirement and thus, without additional stipulations, can interact freely which incorrectly predicts an illicit form to be grammatical. In contrast, the Lowering analysis crucially makes use of the hierarchical coordinate structure to rule out (71), a piece of information that is only present in syntax, prior to linearization.
To sum up the discussion so far, Dagur uniformly displays unexpected order between case and poss morphemes in its nominal domain. The properties of these two morphemes and their relationship with the rest of the nominal domain are explored through the phenomenon of suspended affixation. Specifically, suspending either case or poss from the case-poss sequence is ungrammatical in this language, a pattern that is not observed in other morpheme combinations. The unexpected patterns can be systematically explained with the interaction between the base-generated suspended affixation structure and K-to-Poss Lowering. The Lowering operation manipulates the output of syntax, deriving the case-poss suffix order observed on the surface.

Morphophonological properties of case and poss morphemes
This section focuses on further operations that take place after Lowering which derives the surface morphophonological differences between case and poss morphemes. In terms of morphophonology, case and poss behave differently with regard to how close they are to the stem or the base in Dagur nominals. Specifically, case shows a closer union with the stem or base than poss, displaying what would be traditionally classified as clitic/affix distinctions. First, in Dagur, case morphemes participate in vowel harmony. This is illustrated with In contrast, poss morphemes (73) do not harmonize in any context. This is illustrated with 1pl. poss marker -maan j in (74).
(73) Dagur possessive markers singular plural In nominal phrases that include both case and poss morphemes, we observe that vowel harmony does not extend to poss. This seems to suggest the presence of a prosodic boundary between case and poss, as indicated in (75).
(75) (maani) terɣ -eer ¦ -maan j 1pl.gen vehicle -inst ¦ -1pl.poss 'with our vehicle' Second, in Dagur, case markers triggers idiosyncratic morphological alternations on the stem, whereas poss markers do not. For example, many Mongolic languages have an "unstable" or "fleeting" word-final /n/ in certain word stems that is invisible when the stem stands alone, but reappears when affixes attach to that stem. Dagur has eliminated the unstable /n/ from the declension of regular nouns. However, the unstable /n/ is still preserved in numerals and pronominal declension (Tsumagari 2003), which is triggered by the affixation of case suffixes (76) (e.g., Tsumagari 2003;Engkebatu 1988 stem allomorphy in some contexts, such as the one in (76), and on personal pronouns. 12 Thus, the pattern that emerges is that compared to the poss suffix, the case suffix seems to have a closer relationship with the stem because it can trigger stem allomorphy in some contexts and always participate in vowel harmony. In other words, case morphemes show the properties of what is traditionally classified as affixes, whereas poss morphemes behave more like clitics. 13 As previously mentioned, there are diachronic reasons to regard poss as being recently grammaticalized from post-posed pronominal particles or enclitics. This might provide some intuition for us as to why poss appears to be morphologically and phonologically less associated with the stem compared to case, which has been affixal since very early stages. However, it does not seem to be the case that such diachronic development simply imposes pressure on the linear structure by forcing poss to linearly attach outside of case. As demonstrated in Section 4.3, the analysis that is solely based on linear operations like Local Dislocation makes incorrect predictions about the suspended affixation data. Therefore, there should be other ways to encode the different morphophonological properties between case and poss.
In the next section, I will argue that the morphophonological differences between case and poss are derived from a rebracketing process (i.e., some sort of string-vacuous Local Dislocation) that applies to the structure after linearization. Specifically, K undergoes stringvacuous rebracketing which affixes it onto its host. In contrast, Poss does not undergo this process. It should be emphasized that the discussion so far has shown that Local Dislocation cannot be the sole factor that is causing the linear morpheme order mismatches. However it is reasonable to expect that after Lowering, some sort of linear morphological operations applies and further manipulates the structure. The order of the relevant postsyntactic operations is represented in (78). Due to this order, Lowering feeds rebracketing whenever the relevant context is met, and if Lowering fails to apply, rebracketing will be bled as well.

Linear morphological operations
This component of the analysis utilizes some representations and formalism proposed in Embick & Noyer (2001) and Embick (2007b;. In particular, it distinguishes the following: Based on these definitions, the complex head (80) is an M-Word, while √ ROOT , the terminal nodes Y and X are subwords. Complex head (i.e., M-Words) of this sort can be created mainly via two ways, head movement (assumed here to be part of syntax), and post-syntactic operations which affixes morphemes to each other at PF. I suggest that case morpheme, depending on the given structural condition, is affixed onto its host via either head movement or postsyntactic affixation. Thus, case always gets integrated with its M-Word host. However, poss morpheme does not undergo such process. Therefore, poss is not part of the same M-Word as its host, and simply leans onto the host and be pronounced together with it.
Before moving on to the analysis, some additional assumptions need to be laid out. It is assumed in Embick (2010) and related work that syntactic structures contain only hierarchical information, and an independent linearization procedure is responsible for producing linear representations from the hierarchical structure. As briefly mentioned in previous sections, the binary *-operator indicates linear precedence and adjacency. Therefore, the statement a * b could be read as 'a is left-adjacent to b'. The ⌢ -operator is another binary operator that represents concatenation of M-Words, indicating immediate precedence. For example, the statement a ⌢ b could be read as 'a immediately precedes b'. The ⊕-operator represents the concatenation of subwords, and finally, a general "chaining" process of concatenated elements is assumed, in order to produce the actual linear representation. An illustration of how different levels of constituents are linearized using these operators is given in (81) (a⊕b), (b⊕y) (Embick 2007b) In addition to the mechanism of linearization, I further assume that a relevant rule applies wherever its structural description is met.
(84) Rules Apply (Bobaljik 2017) A rule applies wherever its structural description is met.
We are now ready to discuss possible ways to encode the morphophonological differences between case and poss in the Dagur nominal domain. I got rid of this sentence because it is redundant and expresses the same thing as the previous sentence. To give an overview, I take M-Word to delineate the locality domain for the type of allomorphy observed in (76)-(77). case morpheme, depending on the given structural condition, is affixed onto its host via either head movement or postsyntactic affixation. Thus, case always gets integrated with its M-Word host. However, the poss morpheme does not undergo either process, and is therefore not part of the same M-Word as its host, and simply leans onto the host and be pronounced together with it. Under this analysis, the complex M-Word containing the stem and the case suffix is the domain of allomorphy, which is also relevant to the domain of vowel harmony. M-Word has been suggested to be special for both morphological interactions and phonological processes. With respect to morphological interaction, Bobaljik (2012) suggests that for one morpheme to condition allomorphy for another morpheme, it is necessary for the two to be in the same morphological "word" (i.e., complex X 0 ). Meanwhile, Embick (2010) argues that there are further locality domains delineated by cyclic heads within an M-Word, which may further affect word-internal allomorphy. If Poss is treated as a non-cyclic head, then case and poss should be present in the same cycle, and their different behavior in terms of allomorphy might not be straightforwardly reflected in this regard. However, another way to encode this is that the case suffix is treated as within the same M-Word as its stem, excluding poss. In this way the difference between case and poss becomes structurally transparent. In addition, since case participates in vowel harmony while poss does not, the M-Word domain in (85) seems to be relevant for vowel harmony as well. This connection could be established by relating the M-Word to the phonological word, which in Dagur nominals corresponds to the domain of vowel harmony. Embick raises the generalization that complex heads seem to behave like phonological words or show close phonological connections. This idea is further developed in Shwayder (2015), who argues that non-minimal M-Words (i.e., M-Words that are complex heads instead of terminals) can be related to phonological words in systematic ways. If we take the M-word in (85) to be a phonological word, it would not be surprising that this word delineates the domain for vowel harmony. 14 What is important for the current discussion is that the morphophonological closeness between case and its stem suggests some sort of locality domain at play. In particular, case seems to belong to a locality domain in which it can condition the stem allomorphy, which also happens to be the domain for vowel harmony. I take these facts as an indication that the relevant locality domain in these contexts is the M-Word. Thus, for example, I will take (86a) and (86b) as M-Words.
[ mwd joon -d] what -dat I propose that case belongs to the same M-Word as the stem, however, poss does not belong to that M-Word, but instead loosely attaches, or leans onto the M-Word and be pronounced together with it. This distinction can be derived directly in syntax. Due to the head-final and agglutinative nature of Dagur, it is difficult to determine from the surface form whether the sequence of morphemes collected on the right side are (morpho)syntactically combined into complex heads (either by head movement or postsyntactic operations) or are simply linearly adjacent. 15 According to Shwayder (2015), in a head-final language (assuming right-headedness), the two structures (87) and (88) could have identical surface orders, but distinct morphological structures. The structure in (88) consists of one single M-Word, with the terminals X, Y, Z all being subwords. In contrast, in (87) each head is its own morphological unit.
(87) ZP YP Z X Y a. Linear order: X Y Z b.
Morphological structure: X ⌢ Y ⌢ Z In this way, K is sufficiently close to the stem and can trigger allomorphy. The facts on vowel harmony also naturally follows. Note that the context for string-vacuous rebracketing is created by Lowering. Suppose Lowering does not apply to the hierarchical structure, the scenario would be that poss intervenes between case and its host as in (98). Since free rebracketing may not disrupt the linear order, it would not be able to derive the correct structure in (97). This directly falls out from the general order of postsyntactic operations outlined in (78). Lowering, which applies to hierarchical structure, takes place before linearization. Thus, it will apply to K and Poss whenever the structural context is met, before any other linear morphological rules apply. In other words, due to the general order of postsyntactic operations, Lowering feeds linear rebracketing.
(98) [[ Root * n ] * Poss ] * K ] To sum up the discussion so far, the morphophological differences between case and poss morphemes are encoded in their linear relations. Specifically, through further manipulation of the structure after linearization, case ends up being within the same M-Word as the host, whereas poss does not belong to that domain.

Conclusions
The current analysis provides a case study of the interactions of postsyntactic operations that relates syntactic derivation to surface morphological patterns. The discussion has centered on the Dagur nominal domain, in particular two morphemes case and poss, which exhibit unexpected surface order. The first empirical task of this paper was to determine the morphological and syntactic status of the two morphemes in question. Through investigating the syntactic traits of Dagur nominal phrases, I conclude that Dagur does not project DP. Instead, the noun phrase involves an NP with some functional projections above it. The second empirical task of this paper was to investigate the morphosyntactic traits of the NPs with the exceptional case-poss suffix order, through the lens of suspended affixation. It turns out that coordinated nominals with these suffixes display surprising patterns under suspended affixation. In particular, poss cannot be suspended to the exclusion of case, even though such surface form conforms to the widely observed right edge condition in languages with suspended affixation.
Utilizing the fact that Dagur overt coordinators are sensitive to whether the conjuncts are arguments or not, I have shown that Dagur suspended affixation is most adequately analyzed as a base-generated structure, instead of morpheme ellipsis. Given this, I have proposed that the exceptional morpheme suspension patterns are due to K lowering to Poss postsyntactically. I have also examined further data on the morphophonological distinctions between the case morpheme and the poss morpheme. These surface distinctions are derived by the combination of complex head creation and string-vacuous rebracketing, which places K inside the host M-Word, while excludes poss from that M-Word. The fact that the postsyntactic operations discussed in this paper are sensitive to different types of structure directly falls out from the general architecture of postsyntactic operations. Specifically, Lowering applies in terms of hierarchical structure. Thus, it must precede linearization. The proposed rebracketing operation only applies to linearized structure, not to syntactic hierarchy. Thus, it is ordered after linearization.