This paper proposes an analysis of the movement asymmetry between subjects and non-subjects in Late Archaic Chinese (LAC; 5^th–3^rd century BCE). One instantiation of this asymmetry can be observed in topicalization. Like most modern Sinitic varieties, LAC was an SVO language, as shown in (1a). As I show in section 3, referential subjects in LAC were licensed by moving to a position in the left periphery of the clause. Object topics can also occupy a position in the left periphery, but they do not move to this position, as evidenced by the fact that they have to be resumed by an overt pronoun within VP, as shown in (1b).

The presence of this asymmetry in LAC contrasts starkly with the situation in modern Mandarin, in which object topicalization is derived through movement, which leaves a gap. In (2b), the object has moved to clause-initial position, and no pronoun appears in the base position in VP.

A second asymmetry found in LAC is the employment of separate strategies for forming subject and object relative clauses. Subject relatives are formed by adding the determiner zhe to the end of the clause containing a gap, as in (3a). According to Aldridge (2009; 2013a), zhe is a functional category in the DP layer that serves to bind the gap in [Spec, CP] and form a relative clause. In contrast, LAC relative clauses with VP-internal gaps are required to be nominalized, which is indicated by the fact that the subject is assigned genitive case. The relative pronoun suo also appears between the subject and the VP, as can be seen in (3b).

This paper proposes that the genitive marking on the embedded subject in (3b) plays a crucial role in allowing a relative clause to be formed on a VP-internal position. I propose that DPs undergoing movement in LAC must target case licensing positions. On the assumption that only one structural case feature is available per phase, there is only one such position in any given phase.2 Consequently, movement of the subject to value nominative case generally blocks movement of other DPs into the C/TP layer. But in a non-subject relative clause, the subject can value genitive case with the external determiner. This allows the object to move into the C/TP layer, since it can now be licensed with structural case in the edge of CP. The following section spells out the technical implementation of this proposal within Chomsky’s (2013; 2015) Labeling theory. Sections 3 through 6 then show how the proposal accounts for subject/non-subject movement asymmetries in LAC.

Stated simply, the asymmetry between subject and non-subject movement in LAC can be accounted for by restricting landing sites for moving DPs to positions where structural case is available, as summarized in the preceding paragraph. It may seem surprising, however, that an object DP, which presumably has already valued accusative case, should need to be licensed again. I propose that this is because case valuation alone does not suffice to license an argument; arguments must also be located in positions where Labeling (in the sense of Chomsky 2013; 2015) can take place. In this section, I introduce the approach to Labeling adopted in this paper and lay out how this ensures that bare DPs in LAC can only move to case licensing positions. Chomsky (2013; 2015) proposes that the label of each syntactic object formed through Merge is determined by an algorithm which searches for the closest head within this constituent. When a head merges with a phrase, that head is closer than the one embedded in the phrase and consequently determines the label of the newly created syntactic object.

Among other things, this allows objects to remain in their base positions in VP and not move to a case-checking position in order to be licensed. They can obtain case under c-command and do not need to share a feature with their sister, i.e. the verb, in order for Labeling to take place.

But Labeling is a more complicated process when movement takes place. This is because movement creates a syntactic object consisting of two phrasal categories. For example, in the case of TP in English, T determines the label of the syntactic object formed by merging T with vP, since T is a head. But another mechanism is necessary to label the constituent formed when the subject merges with this TP. For languages like English that have agreement, Chomsky proposes that φ-feature sharing between the subject and its sister is the mechanism which labels this syntactic object.

In cases of wh-movement, it is the [Q] feature shared by C and the interrogative constituent that labels the clause. Note further that [Q] features also tend to be overtly marked in languages that have wh-movement, especially in languages allowing objects to move over subjects to [Spec, CP]. A good example is Tlingit, a language in which a Q-particle merges with an interrogative phrase and projects a QP, as analyzed by Cable (2010). Cable shows that QPs obligatorily move to the left periphery of the clause in Tlingit. I assume that sharing of the [Q] features on QP and C is what labels the clause. I also assume that the wh-component in English interrogative pronouns like who and what is the morphological instantiation of a [Q] feature.

Another type of marking which has been claimed to have consequences for Labeling is morphological case. For example, Japanese has case marking for both subjects and objects. Japanese is also a scrambling language, so although the basic word order is SOV, objects can also move to a position where they precede the subject, as in (7b).

Scrambling of the object over the subject poses a problem for the Labeling Algorithm, since it does not involve any obvious feature-sharing in the landing site. But a solution can be found in Saito’s (2016) proposal for the relationship between morphological case and the Labeling Algorithm. Saito proposes that morphological case has the opposite effect of features like [φ] or [Q] for the purposes of Labeling. He proposes that morphological case makes a DP invisible to the Labeling Algorithm, and this is what allows a scrambled DP to occupy a position where it does not share features with its sister.

Returning now to how movement is licensed in LAC, LAC is a language which lacks overt marking for [Q] and [φ], as well as nominative and accusative case. I propose that the lack of morphological reflexes for [Q] and [φ] makes these features unsuitable to serve as labels. However, case features can serve this purpose in a language that otherwise lacks morphological marking on nominal arguments.3 The intuition behind this proposal comes from the standard Generative assumption that DPs must be case licensed in all languages and in all clause types, i.e. the generalization underlying Chomsky’s (1981) Case Filter. Consequently, I take case feature checking to be the most basic and essential of feature checking operations. Additional conceptual undergirding for the proposal that case features can serve as labels in some languages comes from the common observation that languages lacking overt case marking tend to have rigid word order, while languages with morphological case allow scrambling. I propose that the relative inflexibility in the positioning of nominal arguments in languages lacking morphological case and agreement is a direct consequence of the role of case features in Labeling. Specifically, morphologically unmarked nominal arguments must occupy positions which make their grammatical functions explicit, i.e. positions where their case features are valued. The connection with Labeling is most direct in the case of DPs occupying specifier positions, like the subject. Given that, in the absence of overt case marking, constituents occupying specifier positions must share features with their sisters in order for Labeling to take place, a case feature must be shared if the lack of morphological marking prevents other features from serving this purpose. Consequently, in the absence of other morphological marking, these DPs must occupy case licensing positions. I summarize this proposal in the generalization below.

The proposal in (8) derives the LAC movement asymmetry in the following way. Put simply, the lack of morphological marking in LAC ensures that DPs in LAC can only occupy specifier positions if they are case licensed there, because only case features can serve as labels in the absence of morphological marking. Since the subject typically occupies the position for nominative case sharing, and there is only one case feature per phase, this means that no other DP can move into the C/TP layer as long as the subject needs to be case licensed there. However, if the subject is licensed independently of the case feature on C/T, then a non-subject DP is able to undergo movement into the CP phase edge. I propose that this happens in non-subject relative clauses, where the subject values genitive case under c-command with the external D head. Since the subject does not share the case feature on C, this feature remains available for sharing with a lower argument, thereby allowing a non-subject DP to move into the CP layer in the relative clause. The label for the node dominating the genitive subject, on the other hand, is not a consequence of feature sharing but is simply projected by the subject’s sister node. This is because the overt genitive case marking on the subject DP makes it invisible to the Labeling Algorithm, as per Saito’s (2016) proposal for the relationship between morphological case and Labeling.

As a consequence of this proposal, it appears that a moving object must value case twice, once with transitive v before movement and then again with C after movement. This would be problematic if we assumed the Activation Condition of Chomsky (2000; 2001). Since the case feature of the DP has already been valued by v, the DP should no longer be an active goal for case valuation. However, the Activation Condition has come under fire as an independent principle of syntax by Bošcović (2007). Furthermore, many languages have been argued to allow multiple case valuation, including Niuean (Bejar & Massam 1999), Korean (Yoon 2004), Icelandic (Jónsson 2009), among others. Finally, my approach to the role of case features in Labeling in fact suggests a simple solution to the question of how DPs in languages like LAC can (and sometimes must) value case more than once. This is because the lack of morphological case marking means that the value of the case feature can only be determined positionally. Specifically, bare DPs within the VP are identified as having accusative case, while bare DPs outside of the VP have nominative case. As an extension of (8), then, it can be said that the value of the case feature of a bare DP occupying a specifier position is the case that the DP shares with its sister. Consequently, a bare DP moving to the C/TP layer can only occupy the position for nominative case. Given that abstract case features are valued by phase heads, this proposal can be stated in terms of a phase-based locality condition on case valuation.

(9) also accounts for the cross linguistic generalization that morphological case marking is a prerequisite for scrambling, as noted for Japanese above.4 The overt marking allows a moving object to retain its accusative case feature after moving into a higher phase. The case feature in turn allows the object to occupy a position where it does not share features with it sister without incurring problems for Labeling. But in languages like LAC which lack morphological nominative and accusative case, a moving DP must target a position where it can share features with its sister. Since, according to (8), case features are the only features which can be shared for this purpose in languages lacking morphological marking on nominals, DPs can only move to case licensing positions. And since abstract case values are determined positionally, the value of a DP’s case feature will be set by the head of the phase where the DP is spelled out.5

In the following sections, I show how the proposal made above for the relationship between abstract case and Labeling accounts for several movement asymmetries in LAC. In section 3, I present my analysis of DP movement as licensed by case feature sharing and show how it accounts for the subject/object asymmetry in topicalization in LAC. Section 4 applies this analysis to wh-questions and focus movement and shows that these DPs likewise always target positions where they can be case licensed. In section 5, in preparation for the discussion of nominalized relative clauses, I demonstrate that LAC had a genitive case which was morphologically distinct from the unmarked nominative and accusative cases. Section 6 builds on the existence of morphological genitive case to show how the assignment of this case to an embedded subject allows objects to move to the left periphery in order to form a relative clause.

In this section, I propose an analysis of object topicalization in LAC and show that this is not derived through movement of the topic over the subject. Rather, the topic is base generated in the left periphery and resumed by a pronoun in the clause. I propose that this is because DPs in LAC can only move to positions where they can share a case feature with their sister. Given that there is only one case feature available per phase, then only one DP can be licensed in a movement landing site in any given phase.

To begin, I spell out how subjects are licensed in LAC. The first point to establish is the position where the subject is licensed. I propose that this is typically [Spec, CP] in LAC rather than [Spec, TP] due to the lack of C-T Inheritance. In the framework of C-T Inheritance (Chomsky 2008), the features responsible for licensing nominative arguments are not inherent to T but rather are inherited by T from C. In particular, C passes [uɸ] to T, which licenses the subject and attracts it to the [Spec, TP] subject position. The shared [ɸ] also serves as the label of the newly formed projection. According to Chomsky (2013; 2015), trace positions do not count for the purposes of Labeling, so the subject’s base position in vP is ignored by the Labeling Algorithm. v projects the label for the syntactic object dominating itself and the VP. For clarity, I continue to indicate the base position of the subject and mark this entire projection as vP. If C has a feature driving A’-movement like [Q], this is retained by C, allowing movement over the subject in [Spec, TP] if the clause contains an XP with a matching [uQ] feature, as in cases of object wh-movement. The top-most projection of the clause is then labeled by the [Q] feature.

I follow Richards (2007; 2012) in assuming that C-T Inheritance need not be stipulated but takes place if it is forced. Richards proposes that C-T Inheritance is forced in English because the probe attracting the subject has an unvalued feature which must be spelled out and deleted as soon as it is valued. The reason is to ensure that newly valued features are not transferred to the C-I (Conceptual-Intentional) Interface. If a newly valued feature were not deleted immediately upon valuation, it would be indistinguishable from valued and interpretable features and consequently would be incorrectly transferred to the C-I Interface.

But in languages like Chinese that do not have agreement, there is no reason to assume that C enters the derivation with an unvalued [φ] feature. I propose instead that DPs are licensed by checking a case feature, which does not seek a value and consequently does not need to be inherited by T. This allows the subject to move directly to [Spec, CP]. See Ouali (2006), Goto (2011), Legate (2014), Gallego (2014), Aldridge (2017; 2018), Erlewine (2018), and others for proposals that C-T Inheritance is obviated under certain conditions. Therefore, I propose that the subject moves to [Spec, CP] to check the case feature there, and sharing of this feature between the subject and C is what serves to license the subject and label the CP.

Evidence for the high position of the subject comes from the following contrast between referential and interrogative subjects. A referential subject precedes a high adverb like the modal marker qi, as in (12a), while the non-referential subject follows the same adverb, as in (12b). In section 4.1, I propose that C-T Inheritance does take place in subject wh-questions, triggered by the presence of an uninterpretable focus feature. Inheritance also serves to place the subject in the scope of the [Q] feature on C so it can receive an interrogative interpretation.

The approach to nominal licensing based on Labeling receives support from the asymmetrical behavior exhibited by internal and external argument subjects in terms of their possible surface positions. External argument subjects are required to surface in preverbal position even when they are indefinite. I assume that the movement is obligatory because the subject does not share a feature with its sister vP in its base position and therefore cannot undergo Labeling with it. Consequently, the subject must move to [Spec, CP] where it can share a case feature.

In contrast, indefinite internal argument subjects, like pivots of existential verbs, do not move to [Spec, CP]. Like objects in transitive clauses, these are sister to the verb, and since the verb is a head, it can project a label without sharing features with its sister. Objects and indefinite theme subjects can therefore value case under c-command with v or C and do not need to occupy a case position in order for Labeling to take place. An anonymous reviewer asks why indefinite theme arguments do not – in fact cannot – move out of VP to value case. I assume that this is because they must be in the scope of the existential verb in order for their existence to be asserted. In other words, they must be interpreted within the domain of Existential Closure (Diesing 1992).

Turning now to LAC topicalization, objects do not move to the left periphery; a topic is base generated in clause-initial position and resumed by a pronoun within the clause. I propose that this is because DP movement is licensed by case feature sharing in the landing site. Since the subject must be licensed by the case feature on C/T, the object is unable to move into this phase. Consequently, an object topic can only be base generated in the left periphery and resumed by a pronoun in VP.

However, the presence of a base generated topic in the left periphery introduces a problem for my analysis of movement and DP licensing, given my proposal that only one DP can only undergo Labeling by sharing its case feature with C. Since the subject shares its case feature with C in order to value nominative case, the topic necessarily occupies a position where it cannot undergo Labeling with its sister. I propose that this problem is solved by coindexation with the resumptive pronoun in VP. This allows the topic to be identified with an argument position in the clause. I suggest that this coindexation then serves the same purpose as morphological case-marking on Saito’s (2016) analysis of Labeling. Specifically, these topics are also invisible to the Labeling Algorithm, so the <Case, Case> label is projected from the lower XP projection.

Evidence that the topic is base generated in its surface position comes from the fact that the resumptive pronoun can surface inside islands. (16a) shows the resumptive pronoun as a possessor in the object DP. In (16b), the pronoun is the subject inside a relative clause. Incidentally, embedded subjects are also unable to be topicalized over the matrix subject. (16c) shows that such a topic must also be resumed by a pronoun in embedded subject position. Note that the embedded clause in (16c) is nominalized, so the subject has genitive case.

I assume that LAC base generated topics are parallel to hanging topics in languages like Italian, which have been argued to be base generated in their surface positions rather than being moved from inside the clause (Cinque 1977; 1990; Benincà & Poleto 2004; Frascarelli 2000; 2004; and others). Base generated topics in Italian, unlike moved topics, must be resumed by a pronoun, are not subject to island constraints, and do not show connectivity effects with other elements in the clause. The hanging topic in (17a) is resumed by a clitic pronoun. It also surfaces as a bare DP without the preposition that would be required if the argument surfaced within the TP. The left dislocated topic in in (17b) must have the preposition, and the clitic pronoun is optional. The obligatoriness of overt marking of the moved DP – especially in the absence of resumption – may also be related to Labeling, since the preposition helps to identify the moved object in the same way that case marking does in Japanese.6

My proposal that coindexation between a base generated topic and a resumptive pronoun makes the topic invisible to the LA in the same way as overt case marking does in Saito’s analysis can also be extended to bare adjoined arguments identified with grammatical functions in so-called “pronominal argument” languages (Jelinek 1984). Nominal phrases denoting subjects and objects in such languages are not case marked and also do not occupy designated positions in the clause. According to Baker (1996), they are adjoined in positions in the periphery of the clause and licensed through coindexation with pronominal agreement markers base generated in argument positions within the clause.

Such languages are difficult to account for in the approaches to Labeling put forth by Chomsky (2013; 2015) and Saito (2016). Specifically, since referential nominals do not occupy fixed positions in the clause but are rather adjoined, it cannot be assumed that they move to specific licensing positions where they share features with their sisters. And since they are not overtly case marked, Saito’s analysis does not allow them to be ignored by the Labeling Algorithm. So I propose that the coindexation between the referential nominal and the pronominal agreement marker serves the same function as morphological case marking. Given this function, an extension of Saito’s analysis allows them to be invisible to the Labeling Algorithm. In the same way, I propose that topics adjoined to CP in LAC can also be identified with argument positions in the clause via coindexation and that this coindexation makes them invisible to the Labeling Agorithm.

The lack of topic movement in LAC may also be related to the lack of null object pronouns. Although LAC is a null subject language, null objects are not allowed, even in the absence of an overt topic in the left periphery of the clause. The subject in the answer in (19) is null, but the object is an overt pronoun.

On one view of null pronominalization in Chinese, this asymmetry can be attributed to the lack of movement derived topicalization in LAC. In contrast to LAC, modern Mandarin allows both null subjects and null objects. Huang (1984) proposes that null subjects and objects have a different status, which can be seen by their distribution in embedded clauses. A null embedded subject can refer to the matrix subject, but this is not possible for a null object, as shown in (20a) and (20b), respectively. Huang further shows that if a topic appears to the left of the matrix subject, the null embedded object naturally refers to this topic, as can be seen in (20c).

Huang (1984) concludes that while null subjects can be analyzed as pro, null categories in object position are traces of moved topics. This same analysis can be extended to LAC. The difference between LAC and modern Mandarin is that the variable in object position in LAC cannot be a null category, because LAC does not permit topic movement of an object over the subject. Instead, the resumptive pronoun functions as the variable in object position.

Though accounting for diachronic change is beyond the scope of the current paper, a note is in order regarding the fact that Modern Mandarin differs from LAC in allowing topic movement over the subject, as shown in section 1.

My approach to Labeling does not immediately predict this to be possible, since there is no obvious morphological marking on topics in Modern Mandarin. However, there is evidence that a topic feature may participate in Labeling in Modern Mandarin. Although LAC allows indefinite subjects in clause-initial position, as shown in (13) above, Modern Mandarin does not. As in LAC, an indefinite unaccusative subject whose existence is being asserted remains in its base position in VP, as in (22a). But if the subject moves to preverbal position, it must be interpreted as definite or generic, as indicated by the translation of (22b).

I suggest here that a change may have taken place after the LAC period, which resulted in the addition of a definiteness feature to the case feature responsible for DP movement. I propose that, although a topic feature by itself is insufficient for the purposes of Labeling, the combination of these two features does provide sufficient identification to allow two DPs to undergo Labeling in the C/TP layer.7

In this section, I showed how the lack of object topicalization over the subject in LAC can be accounted for by positing that DP movement in LAC was licensed solely by sharing of a case feature. In the next section, I show how this proposal accounts for movement in wh-questions and focus constructions. In particular, although objects can be dislocated, they always target DP licensing positions within the vP phase.

In the previous section, I proposed that DP movement in LAC is licensed by case feature sharing. As a result, only the subject is able to move to the C/TP layer when it values nominative case, thereby preventing objects from undergoing movement to a topic position in the left periphery. Non-subject topics can only be base generated there and coindexed with a pronoun in the clause.

In this section, I discuss movement of interrogative and focused constituents and show that this movement likewise is licensed by case feature sharing. Consequently, interrogative and focused objects cannot move into the CP phase, but can only target the edge of vP, where they can share a case feature with v.

The preceding two sections demonstrated that objects cannot move into the C/TP layer, because they cannot be licensed there. In section 6, I argue that it is possible for an object to move to the C/TP layer if the subject is licensed under c-command with genitive case. This leaves the case feature on C/T available for the object, allowing the object to move into the C/TP layer. Before discussing how genitive case valuation on the subject accounts for object relative clause formation in LAC, it is first necessary to demonstrate that LAC really does have a distinction between nominative and genitive case. I show in this section that although case distinctions cannot be clearly identified for first and second person pronouns, there is a distinct genitive case for third person pronouns and full DPs. I first show the lack of a clear case distinction in the first and second person pronouns.

Though modern Sinitic languages entirely lack morphological case, many linguists have debated whether a case distinction can be found in earlier Chinese (Ma 1898; Karlgren 1957; Wang 1958; Zhou 1959; Graham 1969; Zhou 1980; Lü 1982; Cui 1989; Zhao 1990; Hong 1991; 1996; Yang and He 1992; Sugita 1993; Pulleyblank 1995; and others). The difficulty lies in the fact that case distinctions are not found in all persons, being neutralized on first and second person pronouns. However, a distinct genitive case can be clearly observed on third person pronouns and full DPs.

I first summarize the syncretism in first and second person pronouns by using first person forms for illustration. The commonly used first person pronouns in LAC are wo and wu. There are two main differences between them, but neither of these is case. Wu is generally singular, as shown in (42a), while wo often has plural reference, referring to a social or political collective like a nation or an army, as in (42b). Wu also occasionally has plural reference but it does not have the collective sense found with wo. Both wo and wu function as possessors in the examples in (42).

The other difference is prosodic. Wo is frequently contrastively focused, as in (43a), while this is never the case for wu. In (43b), wu serves as the subject of the same predicate, the ability modal neng, but there is no contrast in this example.

Kennedy (1956) proposes that wo and wu were a single lexical item differing only in stress. Wu never occurs at the end of a phrase and is never followed by a pause, while this is frequently the case for wo. Consequently, wo is always used when the pronoun is stressed. Wu is also the form most commonly used as a possessor, since it occurs phrase-initially and not phrase-finally. Wo serves as a possessor only when it is stressed or has the collective plural reference. The alternation between wo and wu based on prosody is easily accounted for if we accept the reconstructions of Baxter and Sagart (2014). Wo is reconstructed as a heavy syllable *ŋ^ʕɑjʔ, while wu is reconstructed as a light syllable *ŋ^ʕɑ. See also Feng (2016: 242–243) on the relationship between syllable weight and stress in Archaic Chinese.

Interestingly, a similar pattern can be discerned in the second person pronouns, and this possibility is again mirrored in Baxter and Sagart’s reconstructions. The potentially stressed and frequently plural form is a heavy syllable: 爾 er *nərʔ; while the unstressed, typically singular, form is a light syllable: 而 er *nə. And like the first person pronouns, their distribution does not seem to be determined by case. I leave exploration of the specific analysis of first and second person pronouns for future research. I assume for present purposes merely that the distinction between genitive and non-genitive case is neutralized on these forms. It is very clear, however, on third person DPs.

In contrast to first and second person pronouns, third person pronouns and full DPs do show a distinction between genitive and non-genitive case in LAC. First, there is no evidence for a prosodically based distribution of the type found with first and second person forms. The distribution of genitive third person forms instead seems to have a syntactic basis, being completely confined to use as possessors or subject position in nominalized embedded clauses. Furthermore, there is no phonological basis for positing a derivational relationship of the type just sketched for the first and second person forms. Both of the third person pronouns are light syllables. They share the same rime but differ in onset: zhi 3.NOM/ACC *tə and qi 3.GEN *gə. Furthermore, full nominal possessors and genitive subjects are marked with a syllabic case-marker zhi. In the following discussion, I trace the diachronic development of these forms and provide examples of their use in LAC.

It is generally claimed that Pre-Archaic Chinese (14^th–11^th C. BCE) did not have third person personal pronouns other than demonstratives (Zhou 1959; Wang 1958; Yang and He 1992; Djamouri 1999; Zhang 2001; Wei 2004). The demonstrative zhi commonly occurred in object position or functioned as a possessor in Pre-Archaic Chinese. Other demonstrative pronouns were used in subject position when the subject was focused. As this language was a null subject language (like LAC), subjects were overtly expressed only when stressed.

Zhi developed into a personal pronoun in the Early Archaic Chinese (10^th–6^th C. BCE), continuing to surface as object and possessor, while a second possessor form qi emerged (Zhou 1959; Wang 1958; Qian 2004). By the LAC period, zhi had ceased to function as a possessor, surfacing only in object position when used as a pronominal argument. LAC then had the following third person pronouns: zhi (nominative/accusative) and qi (genitive).

Turning now to examples, zhi could surface in object position, as in (44a), while unstressed subjects were null. A demonstrative surfaced when the subject was focused, as in (44b). It should not be assumed, however, that zhi could only value accusative case. This pronoun also surfaces as the complement of an existential verb, where I assume that it values nominative case with C under c-command. The pronoun in (44c) resumes a topic in clause-initial position.

(45a, b) show possessive third person forms. The pronominal form is shown in (45a). Full DP possessors were also marked with the case marker zhi. This marker is homophonous with the 3^rd person object pronoun and shares the same diachronic origin as a demonstrative pronoun. This pronoun grammaticalized into a neutral (non-deictic) determiner (Djamouri 1999) and, as the head of DP, could function as either a pronoun in argument position or a genitive case marker for a DP in its specifier (Aldridge 2009).16

And the next examples show genitive subjects of nominalized embedded clauses. These two examples are relative clauses.

An anonymous reviewer suggests that the genitive marker on full DP embedded subjects might have been optional, since there are a few examples in which this marker does not appear. But there is clear evidence that genitive marking was not in fact optional. In all texts of the LAC period, pronominal subjects in this environment obligatorily surfaced in their genitive forms. This can only be accounted for if valuation of the case in the syntax was obligatory. Occasional dropping of the genitive marker with full DPs, then, can be accounted for as a post-syntactic prosodic process, due to a stylistic preference to create four-syllable phrases, as in the following example.

Furthermore, genitive case is found on subjects in a variety of other clause types which are uncontroversially nominalized in other languages, such as factives and complements of psych verbs.

Parallel examples can be found in Turkish.

Beginning with Lü (1982), Wang (1958; 1984), and Zhu (1983), many scholars have proposed that embedded clauses like (48) are nominalizations. However, this position is not universally accepted. The alternative approach typically analyzes the genitive case as a marker of subordination (Yang and He 1992; Deng 2015; Mei 2015; Tian & Xu 2016; Yang 2018; and others). However, such an analysis misses the parallel with genitive marking on possessors in noun phrases. It also ignores the crosslinguistic parallel between LAC and languages like Turkish. Furthermore, other types of embedded complement clause do not have genitive subjects. This is the case with assertive embedded clauses like the following. The embedded subjects surface in their bare nominative forms. Therefore, it is difficult to analyze genitive case as a general marker of clausal subordination.

Once again, a parallel can be found in Turkish. Assertive complement clauses are not nominalized and have nominative subjects.

To summarize, it is clear that there is a morphological distinction between genitive and non-genitive case in the third-person paradigm in LAC. I therefore assume that genitive case is available for a DP to value in the syntax. In the next section, I turn to the syntactic role played by genitive case valuation – and its overt realization – in facilitating object extraction.

In this section, I discuss movement of non-subjects to the C/TP layer in LAC. I showed in sections 3 and 4 that only the subject can move out of vP in matrix clauses. This is because DPs can only move to positions where they are licensed by sharing a case feature. Consequently, only one DP can surface in the edge of a given phase. In this section, I show that non-subjects are also able to move to the left periphery of a clause, but this is only possible when the clause is nominalized and the subject values genitive case. In the previous section, I showed that genitive case was morphologically distinguished from nominative and accusative case in LAC. In this section, I propose that the subject in a nominalized relative clause occupies [Spec, CP], as in root clauses, but it values genitive case with an external determiner under c-command. This then allows an operator to also move to the left periphery. I propose that both the subject and the operator move simultaneously into the C/TP layer, the subject targeting [Spec, CP] and the non-subject operator targeting [Spec, TP]. Movement of the object is possible, since the subject values genitive case with the external determiner. Consequently, the case feature on C/T is still available to check with the operator. The genitive case on the subject is also overtly marked, making the subject invisible to the Labeling Algorithm in its landing site, which is necessary since the subject does not share a feature with its sister.

Before presenting the analysis of object relative clauses, I first briefly discuss subject relatives. As mentioned in section 1, LAC subject and object relative clauses have different structural properties. The vast majority of relative clauses in LAC are headless or internally headed. The subject/object asymmetry is also found in externally headed relatives, but for simplicity I limit the discussion here to those without an external head. The distinguishing characteristic of a subject relative clause is the determiner zhe following the embedded clause. (52a) and (52b) show extraction of the external argument in a transitive clause. An anonymous reviewer asks whether extraction of internal argument subjects in unaccusative constructions patterns with subject movement or object movement. (52c) shows that relativization on unaccusative subject position also employs the subject relativization strategy, which helps to show that the DP which checks the case feature on C is the one which is extracted, regardless of its base position.

Adapting the proposal of Aldridge (2009) to the approach I am developing in this paper, subject relatives can be analyzed as finite CPs selected by the determiner zhe. The operator, which is the embedded subject, moves to [Spec, CP] and checks the case feature on embedded C. In this position, it is bound by the determiner to form the relative clause.

Turning to non-subject relative clauses in LAC, these are distinguished from subject relatives in a number of ways. First is the appearance of the morpheme suo between the embedded subject and the VP. The embedded subject also surfaces with genitive case. The gap can be any VP-internal nominal position, e.g. a theme (54a) or a location in (54b).

Suo typically does not relativize on subject position, even in unaccusatives. The gaps in the following examples are locatives.

I propose the following analysis of non-subject relative clauses to account for these characteristics. This analysis is based on Aldridge (2013b) but restates part of that proposal in terms of the analysis developed in this paper. The presence of genitive case on the subject suggests very strongly that object relatives are nominalizations. See Krause (2001), Hale (2002), Aygen (2002), Miyagawa (2008; 2011), Kornfilt and Whitman (2012), and others for proposals that relative clauses with genitive subjects are nominalized in Altaic and various other languages. Aldridge (2013b) proposes that suo is a head merged in v, where it attracts a VP-internal operator to the edge of vP. Alternatively, it is possible to analyze suo as the operator itself, as per Huang (2014) and Huang and Roberts (2017), base generated in the edge of vP binding a gap inside VP. I propose that both the subject and the operator move to the left periphery. The C head enters the derivation with the case feature and also an uninterpretable [uOP] feature. Because both DPs originate in the vP layer, where they do not share a case feature with their sister, they must both undergo movement into the next phase in order to be licensed, which I assume they can do simultaneously. The object is attracted by the [uOP] feature and also must check C’s case feature in order to be licensed. This means that the subject does not check a feature with C, so its landing site cannot be labeled. However, after the determiner is merged, the subject is able to value genitive case with it under c-command. This also circumvents the problem of labeling the topmost layer of the CP, because genitive case is overly marked in LAC, so the subject will be ignored by the Labeling Algorithm, as per Saito’s (2016) analysis of the relationship between morphological case and Labeling. The determiner in (56a) is spelled out overtly as zhe. Non-subject relatives do not require overt determiners, as subject relatives do. The determiner surfaces overtly with a non-subject relative clause when the DP is definite or specific. But it does not bind an argument position in the clause.

Evidence for movement of both the subject and the object suo into the C/TP layer comes from the fact that these elements must precede other vP-external material like aspectual/temporal adverbs and subject-oriented quantifiers, as shown in (57a) and (57b), respectively.

There is also evidence that suo is base merged no higher than the edge of vP. Suo also appears in root clauses to license a gap in the VP in certain constructions. One such environment is existential clauses like the following. In this construction, suo never leaves the vP. Li (2018) analyzes the existential verb you as heading a vP which in turn selects a reduced clausal constituent consisting only of a vP. The base position of the subject and suo are both located in the lower vP edge. In this construction, suo never precedes the subject quantifier jie or aspectual markers, as seen in (58a) and (58b), respectively.

My analysis in turn accounts for the fact that suo cannot move out of vP in clausal contexts like those in (58). This is because it cannot be licensed in the C/TP layer there, because the subject moves to [Spec, CP] to check the case feature on C. This in turn is because this is a matrix clause, so there is no external determiner to license the subject independently of the case feature on C.

It might be countered that the suo phrase in existential constructions is itself a relative clause, in other words a DP selected by the existential verb. This is highly unlikely, however, given the differences between this suo phrase and relative clauses formed on suo. First is the low position of suo. Secondly, an overt subject cannot appear inside the suo phrase in this construction. If this suo phrase were a relative clause, then it would potentially be able to project a DP layer which could license a subject.

My analysis also accounts for the fact that suo only forms relative clauses on VP-internal positions. Because suo is base merged in the vP layer, it can only be coindexed with a VP-internal argument. It may be wondered why suo is not base generated in argument position in VP. This is unlikely to be the case, because of the lack of locality restrictions observed inside the VP. In other words, it is possible for suo to be construed with any DP position within the VP. In (59), the gap construed with suo is the object in an embedded ECM clause, with the embedded subject intervening between the two. Consequently, suo relative clauses do not exhibit the same strict locality within VP observed for object wh-movement in (35) in section 4.1.

One final point to note is that subjects in suo relatives are indeed located inside the embedded CP and do not move to [Spec, DP] to value their genitive case. First, a temporal adverb can precede the genitive subject. Miyagawa (2008) uses similar evidence to argue that genitive subjects in Japanese relative clauses likewise do not raise to [Spec, DP].

Secondly, the genitive subject clearly occupies a position structurally lower than the external determiner zhe. As pointed out above, this determiner can be added to an object relative clause in order to make the nominal definite or specific. In the following examples, two object relative clauses are coordinated. One determiner follows both clauses, and both embedded subjects have genitive case.

In this section, I have argued for an analysis of non-subject relativization which allows a non-subject operator to move to the C/TP layer in the embedded clause and check the case feature there. This is possible because the embedded subject values genitive case with the external determiner and consequently need not be licensed by or undergo Labeling with the embedded C.

In this paper, I proposed that several movement asymmetries between subjects and non-subjects in Late Archaic Chinese are accounted for by positing that DP movement in this language is licensed solely by sharing a case feature in the landing site. Specifically, sharing of a case feature is what labels the landing site for a dislocated nominal argument. Because only the case feature can license movement of a DP, only one DP can move into a given phase, with the result that subjects move to [Spec, CP], but objects can generally move no higher than the edge of vP. This proposal accounts for the requirement that object topics be resumed by pronouns in VP. I also extended this analysis to focus constructions and showed that, although focused and interrogative objects undergo dislocation, they never vacate the vP.

In order for a constituent other than the subject to move to the C/TP layer, the subject has to be licensed independently of the case feature on C/T. Thus, non-subject relative clauses are nominalized, which allows the subject to value genitive case with an external determiner. Because genitive case is overtly marked in LAC, the genitive subject is invisible to the Labeling Algorithm, so it can occupy a position where it does not share a feature with it sister. This leaves the case feature on C/T in the embedded clause available to be shared with the object.

In section 2, I also suggested an underlying explanation for why DP movement is licensed solely by checking a case feature in some languages. I proposed that features which can be shared for the purposes of Labeling typically must have an overt morphological reflex. However, in languages with highly impoverished nominal morphology, case features can be shared for this purpose. This is because, beginning with Chomsky’s (1981) adoption of the Case Filter, it is standardly assumed that DPs require licensing in all clause types in all languages. This in turn affords a fundamental and essential role to DP licensing. In other words, if no other feature is available for labeling the positions occupied by nominal arguments (due to the lack of overt realization), then case features can serve this purpose. This in turn yields languages like LAC in which case positions are the only syntactic positions which can serve as landing sites for DP movement.