2.1 Puyuma morphosyntax basics

Puyuma is a severely endangered Austronesian language spoken in southeastern Taiwan with less than 1,500 speakers (UNESCO 2010). Prior to this study, its RTO construction was reported in two reference grammars (Huang 2000; Teng 2007), both of which described it as a raising construction. The specific constraints and properties of this construction, however, have remained underanalyzed. Before entering into the discussion, I present basic facts of Puyuma relevant to the analysis of RTO.

As a typical Philippine-type language, Puyuma is predicate-initial, and possesses a four-way voice system with an elaborate argument-marking mechanism. The mapping between voice-marking and the argument-marking pattern in the language is presented in Table 1.

As seen above, in Puyuma, when a clause is marked in Actor voice (AV) (-em-), Pivot-marking falls on the external argument, with the internal argument Accusative-marked (10a).³ When a clause is in Patient voice (PV) (-aw), Pivot-marking falls on the internal argument, with the external argument Genitive-marked (10b). When a clause is in Locative voice (LV) (-ay) or Circumstantial voice (CV) (-anay), Pivot-marking falls on the locative phrase and the benefactive or instrumental phrase, respectively. In either voice, the external argument carries Genitive-marking, and the internal argument carries Accusative-marking, as seen in (10c)–(d).

There is a noteworthy exception to this pattern. Like many other Philippine-type languages, Puyuma has a number of verbs that are morphologically marked as LV (-ay) but take a PV argument structure. In such cases, the LV-marked verb selects no locative phrase, but employs a Pivot-marked internal argument, as seen in (11a)–(b). Such verbs are glossed as “LV[PV]” throughout the paper.

The argument-marking system of Puyuma is presented in Table 2. As seen below, case-markers in the language are portmanteau in function, specifying both the case status of a phrase and the number and definiteness of its referent. In the dialect (Nanwang) investigated in this paper, Genitive and Accusative case have undergone morphological syncretism (Teng 2009). A Genitive/Accusative distinction is nevertheless evident by the presence or absence of a Genitive proclitic that crossreferences the Genitive phrase. As seen in (12a)–(b), the external argument of a non-AV clause is obligatorily present as a pronominal proclitic, which crossreferences a Genitive-marked proper name (12a). If the external argument is a pronoun, it appears merely as a proclitic. Accusative phrases, on the other hand, are not cross-referenced by a proclitic (e.g., kana kuraw ‘the fish’ in (12b)), which are therefore distinguished from the Genitive phrases. For the sake of clarity, I maintain a Genitive/Accusative distinction throughout the glosses in this paper.

The examples below illustrate how the definiteness distinction in argument-marking is manifested. In (13a), both the external and internal arguments bear an indefinite marker, whereas in (13b), both arguments are definite-marked. As both examples are in AV, there is no proclitic on the verb corresponding to the external argument bangsaran ‘young man’.

Finally, it is important to note that Puyuma imposes an A’-extraction restriction commonly found in Philippine-type languages, in which only the “Pivot”-marked phrase is eligible for A’-extraction. This constraint is known as “Pivot-only.” As exemplified in (14), when a clause is AV-marked, only the external argument, i.e., the Pivot, is accessible to relativization (14a). The internal argument cannot be relativized (14b), as it is not the Pivot of the AV-clause:

With this background in mind, I present the basic facts of Puyuma RTO in the following subsections.

2.2 The finite CP analysis of RTO complements

I begin by clarifying the size and finiteness of the complement clause in the Puyuma RTO construction.

Puyuma RTO is associated with verbs that select a finite CP complement, and is most commonly observed with knowledge and perception verbs (e.g., ‘see’, ‘know’, ‘hear’, ‘dream’, ‘pray’, ‘fear’, ‘forget’, ‘miss’, ‘like’). Most languages in which RTO constructions are attested allow only a limited number of verbs to form an RTO structure (e.g., Japanese: Kuno 1976; Tanaka 2002; Passamaquoddy: Bruening 2001; Tsez: Polinsky & Potsdam 2001; Korean: Yoon 2007; Romanian: Alboiu & Hill 2013). In Puyuma, on the other hand, RTO is fully productive with CP-taking verbs.

The finite CP analysis of the RTO complement is evidenced by three major differences from infinitives. First, a complementizer dra is obligatorily present in the complement of RTO (15a), just as in non-raising sentences selected by the same matrix verb, as in (15b). In these examples I indicate the XP’s thematic equivalent in the embedded clause as “e.c.” (empty category), followed by a parenthesis that indicates its case status, as seen in (15a):

In contrast, infinitives in Puyuma do not allow a complementizer, as in (16a)–(b):

Second, the RTO complement is compatible with all types of voice markers (17a)–(d), as opposed to infinitives, which impose an “AV-only” constraint in voice-marking, whereby Actor voice is the only available voice marker on infinitival verbs (18a)–(b).

Third, the complement of RTO is fully compatible with all types of aspect markers (19), as opposed to infinitives, which cannot host aspect markers (20):

Given the observations above, it can be concluded that the complements of Puyuma RTO are finite CPs.

2.3 The matrix behaviors of the XP

The XP in Puyuma RTO shows matrix object-like behaviors with respect to case-marking and binding. As seen below, an XP shows no case connectivity effect with the embedded clause, and its case-marking is fully dependent on the voice-marking of the matrix verb. When the matrix verb is in AV, the XP must bear Accusative-marking (21a), like a normal object of an AV verb (21b). When the matrix verb is in PV or LV, the XP obligatorily bears Pivot-marking (22a), like a normal object of PV verbs (22b):⁴

As can also be seen in the data above, Puyuma RTO imposes no voice-marking constraint on either the matrix or embedded verb, except for the fact that Circumstantial voice (as well as true Locative voice) is not an available voice type for the matrix verb of RTO.⁵

The matrix behavior of the XP is also manifested in binding. As seen below, at the “raised” position, a reflexive XP must be bound by a matrix antecedent, and it cannot be bound by the embedded external argument—which c-commands the XP’s thematic equivalent in the embedded clause. Therefore, in the RTO sentence (23a), the XP kantaaw ‘himself’ can only be interpreted as bound by the matrix external argument ‘Siber’, whereas in the non-raising sentence (23b), the thematic equivalent of the XP is locally bound by the embedded external argument, ‘Isaw’. Therefore, the RTO sentence (23a) bears a different reading from its non-raising counterpart (23b).

A similar effect is observed with RTO sentences that involve an XP containing a possessor. In such cases, only the matrix external argument and not the embedded external argument can bind into the XP and be interpreted as its possessor. Therefore, in the RTO sentence (24a), the XP tu=ngiyaw ‘her cat’ is most naturally interpreted so that the possessor is coindexed with the matrix subject Akang, whereas in a non-raising sentence (24b), tu=ngiyaw may be interpreted as the possessor referring to either the embedded external argument or the matrix external argument, resulting in a different reading from (24a).⁷

As shown above, the XP in Puyuma RTO shows matrix object behaviors in both case-marking and binding.

2.4 The definiteness constraint on the XP

As introduced in Section 1, Philippine-type Austronesian languages commonly impose a “Pivot-only” constraint on the XP in RTO, which prevents non-A’-extractable phrases from participating in raising. In Puyuma RTO, however, this constraint is absent. As seen in (25), phrases with different case status and grammatical relations may serve as an XP. In (25a), the XP ‘Sayki’ is thematically identified with the Genitive-marked external argument of the embedded clause, whereas in (25b), the XP nadru walak ‘that child’ is thematically linked to the Accusative object of the embedded clause. In (25c), the XP ‘Tripul’ is thematically identified with a locative adjunct:

While a “Pivot-only” constraint on the XP is absent and the construction imposes no restriction on the grammatical relation of the XP, the XP in Puyuma RTO must be definite or be interpreted as generic if it bears an indefinite marking. As seen in (26a) and (27a), a phrase can be marked as either definite or indefinite in its theta-position ((26a), (27a)), whereas an XP must be definite-marked at the “raised” position in an RTO sentence ((26b), (27b)):

Exceptions to this constraint are found only when the XP has a generic reading. In such cases, the indefinite-marked XP may surface at the “raised” position, followed by a CP that denotes a proposition about the generic XP, as in (28a)–(b):

Besides such cases, the XPs in Puyuma RTO must be definite. This constraint serves as an important piece of evidence for understanding the nature of this construction, which will be discussed in Sections 4–5.

3.1 Island immunity

The non-movement status of the XP is first indicated by the absence of island effects in the dependency between the XP and its correspondent in the embedded clause. As seen in (30), an XP can be identified with an empty category embedded inside a complex NP. That (30b) is wellformed thus suggests that the XP does not undergo A’-movement from the embedded clause.

Contra the observation from RTO, relativization (pseudo-clefting) in Puyuma is sensitive to island effects, whereby a phrase embedded inside a complex NP cannot be extracted (31a)–(b). This suggests that, unlike RTO, relativization in Puyuma involves A’-extraction.

Consistent with its immunity to complex NP islands, the XP in Puyuma RTO is insensitive to adjunct islands. As seen in (32), the XP ‘Isaw’, whose thematic equivalent is embedded inside an adjunct clause (32a), is eligible to surface at the “raised” position to form an RTO construction (32b). The well-formedness of this sentence again suggests that the XP does not undergo A’-movement from its theta-position.

This analysis is additionally confirmed with the data in (33), which shows that A’-extraction (pseudo-clefting) in Puyuma is sensitive to adjunct islands, as in (33):

3.2 The absence of reconstruction effects

A second argument for the non-movement analysis of the XP comes from its lack of reconstruction effects. As shown in the data below, an XP cannot contain a pronoun that is interpreted as a variable bound by the embedded external argument, hence the unavailability of a distributed reading between the XP and an embedded quantifier external argument. Therefore, in (34b), a bound variable reading is not available between the embedded quantifier external argument bulraybulrayan driya ‘every girl’ and the XP kantu arepu ‘her hair’. Likewise, a bound variable reading is not available between the embedded external argument suwan driya ‘every dog’ and the XP tu ikur ‘its tail’ in (35b):

The in-situ status of the Puyuma XPs is further evidenced by their difference in behavior from those in movement-type RTO constructions, which show reconstruction effects. According to previous studies, in instances of RTO whose XPs are sensitive to island effects, the XP can be interpreted as a variable bound by the embedded subject, suggesting that the XP undergoes movement from the embedded clause (see, e.g., Japanese RTO: Tanaka 2001; Passamaquoddy RTO: Bruening 2001; Romanian RTO: Alboiu & Hill 2013). In Passamaquoddy (Bruening 2001: 6), for instance, an XP can be bound by a quantifier external argument and interpreted as a variable (36b), suggesting that it undergoes A’-movement from the embedded clause:

Given the observations above, I conclude that Puyuma RTO is best analyzed as containing an XP base-generated in its spell-out position. Several important traits of this construction follow consistently from this analysis: first, the XP shows no case connectivity effect, indicating that the XP does not undergo movement from the embedded clause, and second, the XP is insensitive to a “Pivot-only” constraint, indicating the absence of A’-movement in this construction.

3.3 The relationship between the XP and the embedded CP

Having presented a non-movement analysis for the XP in Puyuma RTO, I address an important question: how is the relation between the base-generated XP and the embedded CP established? In this section, I demonstrate that the relation between the two is formed by the aboutness condition.

We have seen in Section 1 that an XP can be identified with an adjunct of the embedded clause. Two examples of this type are presented below, whereby the XP is identified with a temporal (37a) or locative (37b) adjunct of the embedded CP:

These examples suggests that the presence of a gap might not be necessary for the complement clause of Puyuma RTO, indicating that the relation between the XP and the embedded CP in this construction cannot be attributed to a conventional proleptic analysis—according to which the XP is thematically linked to the embedded clause through coindexation with an embedded pronoun, as in (38) (Higgins 1981; Panhuis 1984; Massam 1985; Davies 2005; Salzmann to appear):

The data below shows further evidence that the relation between the XP and the CP in Puyuma RTO is distinct from that of a proleptic construction. In both sentences the embedded clause is gapless, but denotes a proposition relevant to the XP. The well-formedness of both examples (39)–(40) suggests that the XP-CP relation in Puyuma RTO may be established simply through the pragmatics.

I argue that the relation between the XP and the embedded CP in Puyuma RTO is best analyzed as formed by the aboutness condition (Reinhart 1981; Gundel 1985; Lambrecht 1994; Jacobs 2001) (41), whereby the XP is essentially an aboutness topic of the embedded clause.

As an aboutness relation between a topic and a CP can be established either through coindexation with an embedded pronoun or simply through the pragmatics, the embedded CP in a Puyuma RTO sentence may but need not contain a pronoun. This analysis is illustrated in (42):

If this proposal is on the right track, Puyuma RTO is predicted to show two characteristics: (i) the XP should exhibit topic properties, and (ii) the RTO complement must be pragmatically connected to the referent of the XP to satisfy the aboutness condition. In this subsection, I discuss evidence for the second prediction. The topic properties of the XP will be discussed in Section 4.

Consistent with the proposal in (42), example (43) shows that failure to establish an aboutness relation between the XP and the CP yields semantic infelicity and makes an RTO sentence unacceptable. As seen below, when the embedded CP in RTO does not contain a pronoun coindexed with the XP, the content of the CP must be pragmatically connected to the XP. Therefore, replacing the embedded Pivot etruk ‘carp’ with ladru ‘mango’—which is not pragmatically connected to the XP ‘fish’—makes the RTO sentence unacceptable.

In line with the observation from (43), example (44) shows that an RTO sentence is infelicitous if the pragmatic connection between the XP and the CP is missing. Without context, the sentence in (44) is unacceptable, as the XP ‘Sayki’ is not pragmatically linked to the content of the embedded CP “Siber bought a car”. However, a Puyuma speaker I consulted noted that (44) is potentially acceptable if the XP ‘Sayki’ and the embedded Pivot ‘Siber’ have a certain relationship that is known by both the speaker and the addressee. For instance, if the XP ‘Sayki’ refers to the wife of Siber, (44) is acceptable as a propositional sentence about Sayki. This interpretation lends further support to the current analysis, that the relation between the XP and the CP is established through the aboutness condition.

Additional evidence for this analysis is shown in the data below: In order to establish the aboutness relation between the XP and the CP, the XP ‘Atrung’ in (45) must be interpreted as the possessor of the embedded possessive phrase ‘her house’, as the XP cannot be pragmatically linked to any other argument within the embedded CP. Therefore, although in simple clauses (e.g., (46)), both ‘Senten’ and ‘Labu’ are a potential binder of the possessive phrase ‘her house’, in the RTO sentence (45), only the XP (‘Atrung’) can be interpreted as the possessor of the pronominal phrase ‘her house’ in order for the aboutness relation between the XP and the CP to be established.¹⁰

Given the observations above, I conclude that the relation between the XP and the CP in Puyuma RTO is best analyzed as established through the aboutness condition.

4.1 The structural relation between the XP and the embedded CP

Support for the XP being internal to the embedded CP comes from two independent pieces of evidence. First, an XP cannot be separated from the embedded CP in linear order, as seen in (48):¹¹

In contrast, a genuine matrix object can be separated from a following complement by the same temporal adverb adaman ‘yesterday’, as in (49):¹²

The difference between (48) and (49) suggests that the XP in Puyuma RTO must not be base-generated in a matrix object position, but is situated in the embedded left periphery. This analysis is reinforced by an observation from Madurese proleptic construction (Davies 2005), where an XP is free to surface to the left of a matrix temporal adverb, as in (50):

Despite the lack of in-depth analysis on the structural position of the XP in proleptic constructions, previous work has generally assumed it to be base-generated in a direct object position and be independent of the embedded clause (see, e.g., Davies 2005, which proposes the XP to be base-generated at [Spec, AgrOP]; see also Kurniawan 2012 and Salzmann to appear for a similar assumption). The difference in acceptability of an intervening adverb between the XP and the CP between Madurese prolepsis and Puyuma RTO therefore suggests that the latter employs a structure distinct from prolepsis.

The second argument for the current analysis—that the XP in Puyuma RTO is inside the embedded CP—comes from the case-marking of the XP. As shown in all data presented in this paper, in Puyuma, there can only be one (particular) phrase that is eligible for Pivot-marking within each finite CP. This “Pivot as unique constraint” is stated in (51) and illustrated with the data below in (52)–(53). As (52) shows, when a verb is PV-marked, the Theme DP of the sentence must be Pivot-marked.

When a non-AV marked verb selects two objects, only one of the two is eligible for Pivot-marking. This is seen in (53): when a ditransitive verb is in PV, only the Recipient can be Pivot-marked (53a); when the verb is in CV, only the Theme can be Pivot-marked (53b). A “double-Pivot-marking” pattern results in ungrammaticality.

As introduced in Section 2, other than argument-marking the Pivot status of a constituent is manifested also by its accessibility to A’-extraction. As shown in (54), only when the internal argument of a verb is in Pivot status can it undergo A’-extraction. Therefore, only when the verb ‘forget’ is in PV can the object ‘key’ be pseudo-clefted:

When a knowledge/perception verb selects a CP complement, as in (55), the Pivot status of the CP is not morphologically realized, but is nevertheless manifested in the CP’s accessibility to A’-extraction. The special constraint has been observed in various Philippine-type Austronesian languages, and has led to the claim that CPs are Case-licensed in these languages, just as DPs are (e.g., Chung 1994; Pearson 2005; Rackowski & Richards 2005; Chen & Fukuda 2016). See the data below from Puyuma (56) and Tagalog (57):

Given the extraction facts presented above, we can conclude that a CP complement bears Pivot status when it is selected by a PV-marked verb, as in (58):

Building on this generalization, consider again the RTO sentence in (59):

Given that Pivot-marking must be unique within a single clause (see the “Pivot as unique” constraint in (51)), the fact that the XP in the RTO sentence (59) bears obligatory Pivot-marking thus suggests that the XP and the CP must form a single constituent. If they do not, the Pivot status of both the CP and the XP would suggest an argument-marking pattern that is otherwise disallowed in Puyuma.

Further support for this proposal comes from the data below in (60), which shows that A’-extraction out of the embedded CP is possible in an RTO sentence.¹⁴ The fact that an object can be extracted from the embedded CP in this sentence indicates that the embedded CP must be in Pivot status.

Given the observations above, I conclude that the XP’s apparent matrix-object behavior in case-marking is best analyzed as reflecting the case assigned to the CP. This analysis is illustrated in (61):

Below I provide an account for why the XP—as a left-dislocated phrase in the embedded [Spec CP]—manifests matrix-object-like case-marking. Following recent proposals by van Koppen et al. (2016) and Pearson (2018), I argue that a left-dislocated phrase can have its morphological case determined “from above” by a c-commanding head, as long as it is in a local structural relation to that head. Following Pesetsky (2013), I assume that morphological case is the reflex of categorial feature copying in the syntax, where feature copying occurs when a head merges with a dependent. For example, when a head F selects a complement, the lexical category feature of F is copied onto all the terms of that complement. If the complement contains a DP, the feature copied from F may be realized on that DP—which is expressed through insertion of a particular morphological case when the complement is spelled out.

Under this approach, in Puyuma RTO constructions, merger of the matrix V with the embedded CP causes V’s categorial feature to be copied onto every term of the CP, including the XP, which is base-generated in [Spec CP].¹⁵ Since the XP is a DP and is locally c-commanded by the matrix V, it will realize this categorial feature in the form of morphological case. Following the existing accusative approaches to Philippine-type Austronesian languages (e.g., Chung 1994; Pearson 2001; 2005; Rackowski 2002; Rackowski & Richards 2005; Chen 2016, 2017), I assume that Puyuma exhibits a nominative-accusative Case system, and that the marker “Pivot” is a topic marker that overrides morphological case, which falls on the external argument in AV clauses and the internal argument in PV clauses, as illustrated in Table 3. Under this analysis, the CP complement of a knowledge/perception verb always receives (abstract) accusative Case, which is spell-out as morphological case and copied onto the XP. When the CP complement is in Pivot status, however, the accusative case on the XP is overridden by “Pivot”-marking, resulting in ECM-like phenomenon in Puyuma RTO constructions.

In sum, given evidence from word order restrictions on the XP (see (48)) and the “Pivot as unique” constraint discussed above (see (51) and its associated discussion), we can conclude that the XP in Puyuma RTO is best analyzed as a dislocated phrase base-generated in the embedded left periphery.

4.2 Accounting for the binding facts in Puyuma RTO

Having addressed the structural relation between the XP and the CP, an important question needs to be answered: if the XP is indeed situated in the embedded left periphery (62), how can it manifest matrix object behaviors in binding, as discussed in 2.2? In this subsection, I show that these apparent matrix behaviors of the XP in fact follow from the present analysis.

As the XP is base-generated at the embedded left edge, it is predicted to be unable to be bound by the embedded external argument, as is indeed observed in the previous data (23)–(24) in Section 2.2. On the other hand, given that anaphoric binding in Puyuma can cross a clausal boundary—as seen in (63a)–(b)—we expect that an XP can be bound by a coreferential matrix external argument, as it is the closest antecedent of the XP. This prediction is indeed borne out with the observations in the previous examples (23b) and (24b), repeated below in (64a)–(b):

Given these observations, it can be concluded that there is no conflict between the current analysis of the XP and its apparent matrix object-like behaviors.

5.1 Hanging topics in Puyuma

In this subsection, I summarize basic traits of hanging topics in Puyuma. Puyuma has a sentence-initial position that can be filled by three types of phrases: (i) referential definite DPs, (ii) indefinite DPs that bear a generic reading, and (iii) adjuncts that embed a definite DP. Indefinite DPs cannot occupy this position (66b), unless they have a generic reading, as in (66c)–(d). Adjuncts that contain a definite DP and an adverbial clause may also surface at this position, as in (66e):

In constructions like (66a)–(e), the left-dislocated phrase (henceforth YP) is followed by a particle i and a pause. Phrases eligible to occupy the sentence-initial position show characteristics typical of a base-generated phrase. First, similar to what is observed with the XP in RTO, failure to establish an aboutness relation between a YP and the root clause results in semantic infelicity. This is seen in (67)–(68): in order for the sentence (67) to be acceptable, the possessive phrase ‘her house’ embedded inside an adjunct must be interpreted as being possessed by the YP ‘Senten’. Example (68) further shows that a YP must be pragmatically associated with the root clause. Thus, replacing the embedded Pivot pulrikudrakudran ‘chrysanthemums’ with pangudral ‘pineapples’—which is not pragmatically linked to the hanging topic ‘flowers’—results in semantic infelicity:

Further, several pieces of evidence suggest that the YP does not undergo A’-movement from the root clause.¹⁶ First, similar to the XP in RTO, the YP is insensitive to a “Pivot-only” constraint, and can be identified with phrases of all types of case status and thematic roles. As seen in (69), a YP can be identified with a Genitive external argument (69a), an Accusative internal argument (69b), or a locative adjunct (69c):

Second, the dependency between a YP and its correspondent in the root clause does not show island effects. As seen in (70), a YP can be identified with an external argument embedded inside a complex NP island, indicating that it does not A’-move from its theta-position.

Consistent with the observations above, the YP shows no reconstruction effects. As seen in the data below, a pronominal YP (‘her child’) cannot be interpreted as a variable bound by a quantifier external argument (‘every mother’), as in (71b):

Finally, the YP shows no case connectivity effects with the root clause. As seen in (72), a phrase that occupies the topic position must carry the morphological marking ‘Pivot’, regardless of its case status in the root clause:¹⁷

Given the observations above, I argue that the YP exhibits the hallmarks of a hanging topic/aboutness topic, which is base-generated extra-sententially and pragmatically connected to a clause via the aboutness condition (e.g., Aissen 1992; Anagnostopoulou 1997; Zeller 2009; Miyagawa to appear). The observations that a YP must be either definite-marked or bear a generic reading follows directly from this analysis, as aboutness topics are commonly observed to be subject to a constraint that they must be definite or generic (see, e.g., Reinhart 1982; Lyons 1999; Krifka 2001; Cruschina 2016).

In what follows, I discuss how this hanging topic construction sheds light on the nature of Puyuma RTO.

5.2 Puyuma RTO as an embedded topic construction

As foreshadowed in the preceding subsection, the XP in Puyuma RTO shows behaviors parallel to hanging topics (the YPs). Both exhibit the hallmarks of a base-generated phrase, evidenced by their insensitivity to a “Pivot-only” constraint, as well as their immunity to islands, lack of reconstruction effects, and absence of case connectivity. Furthermore, both are subject to a definiteness constraint, except for cases where they bear a generic reading. These similarities are summarized in Table 4.

Given Table 4, I argue that Puyuma RTO is best analyzed as an embedded topic construction that contains an aboutness topic base-generated at the embedded [Spec CP], whose structure has been discussed previously in Section 4.1. The current analysis is in concord with Massam’s (1985) proposal that apparent cases of RTO constructions may contain an XP that functions as an embedded topic, which may or may not undergo further ECM-movement into a matrix A-position (Massam 1985: 115–23). The non-movement analysis of the XP further concurs with a recent proposal in Landau (2011), that the relation between a gapless propositional CP and a hanging topic can be established via the aboutness condition.

The aboutness topic analysis of the XP offers a compelling account for several important traits of Puyuma RTO: (i) the XPs are subject to a definiteness constraint (2.3) while are insensitive to the constraints on A’-extraction (3.1–3.2), and (ii) the construction need not contain a gap in the embedded clause (3.3). Finally, the observation that Puyuma RTO is fully productive with CP-taking verbs follows from this analysis, as verbs which select a CP complement presumably cannot restrict whether that CP includes a topic.

An anonymous reviewer noted that if the present analysis is on the right track, the embedded gap in Puyuma RTO (whenever present) should be fillable with an overt pronoun, as should the gap in a hanging topic construction. This prediction is indeed borne out. As shown in the preceding discussion, both an XP and a hanging topic (YP) can be coindexed with a possessive phrase (e.g., (66b), (70)). Furthermore, as introduced in Section 2.1, when the embedded clause of an RTO construction is non-AV-marked, a Genitive proclitic is obligatorily present on the embedded verb, which crossreferences the XP, as seen previously in (21a) and (22a). The same observation applies to hanging topic constructions, as seen above in (69a) and (70). According to primary fieldwork, only when the embedded clause in an RTO sentence is AV-marked is an overt pronoun dispreferred (73a). A parallel observation is found with hanging topic constructions (73b), where the spell-out of an overt pronoun in the theta-position of the topic phrase is disfavored:

A final question regarding the topic analysis of the XP concerns the observation that unlike the YPs in hanging topic constructions (74a), a presence of the particle i following the XP in Puyuma RTO is considered redundant and disfavored by speakers (74b):

I remain agnostic about the nature of this asymmetry, and tentatively propose that the particle i is the spell-out of the functional head that introduces the aboutness topic at the C domain. In matrix environment, this functional head is always spelled out, as the matrix complementizer is not morphologically realized in Puyuma. In embedded environment, on the other hand, this functional head is preferred to be null, as spelling out both this head and the embedded complementizer dra—which is obligatorily spelled out in all sentences with a finite embedded clause—results in two adjacent functional words, which is disprefered in Puyuma grammar. This possible analysis requires further investigation.

6.1 The microvariation in non-movement type RTO constructions

I have demonstrated in the preceding sections that Puyuma RTO contains a base-generated topic whose relation with the finite embedded CP need not be established via an embedded gap. As briefly discussed in Section 3.3, the acceptability of a gapless embedded CP in Puyuma RTO suggests that this construction employs a mechanism different from similar constructions found in three other Austronesian languages, Madurese, Sundanese, and Tagalog (Davies 2005; Law 2011; Kurniawan 2012), despite their superficial resemblances in Table 5.

As seen above, RTO in all four languages employs a finite CP complement and an XP that shows the hallmarks of a base-generated phrase, manifested in its insensitivity to islands, lack of reconstruction effects, and case connectivity. However, a closer look at these four constructions reveals interesting microvariation. First, in Puyuma, a pro in the embedded clause is not necessary, whereas in Madurese, Sundanese, and Tagalog a pro is required (Davies 2005; Gerassimova & Sells 2008; Law 2011; Kurniawan 2012). Second, Puyuma, Madurese, and Sundanese do not impose a “Subject/Pivot-only” constraint on the XP, while Tagalog does (Kroeger 1993; Gerassimova & Sells 2008). In addition, RTO in the first three languages shows full productivity with CP-taking verbs, whereas Tagalog RTO is restricted to a small number of knowledge/perception verbs (Law 2011; primary data). In what follows, I discuss the ways in which the differences in these constructions illuminate the strategies available for establishing a relation between a left-dislocated XP and a CP.

The RTO constructions in Madurese, Sundanese, and Tagalog have previously been analyzed as instances of prolepsis (Davies 2005; Law 2011; Kurniawan 2012), which is defined as follows:

Along the line of previous work, Salzmann (to appear) has explicitly argued that a coreferential element in the embedded clause is obligatory in a proleptic construction. This constraint is illustrated with the English examples (76a)–(b) (Salzmann to appear: 1):

Consistent with this definition, the RTO constructions in Madurese, Sundanese, and Tagalog have each been described as requiring an embedded resumptive element coindexed with the XP. In Madurese and Sundanese, the resumptive element can be manifested as an overt pronoun (77a)–(b) (Davies 2005; Kurniawan 2012), whereas in Tagalog, it is usually a null pro (Law 2011; primary data), as in (77c):

The differences between these three constructions and Puyuma RTO suggests that the relationship between a base-generated left-dislocated phrase and a CP can be established through at least two strategies: coindexation, as employed by Madurese, Sundanese, and Tagalog (78a), and an aboutness condition, as employed by Puyuma (78b):

Finally, the case of Tagalog RTO further suggests that languages may employ an independent constraint to restrict the coindexation relation between a left-dislocated phrase and a CP. As seen below, an XP in Tagalog RTO must be identified with an embedded Pivot phrase, while this is not the case in Madurese and Sundanese RTO:

In the following subsection, I turn to the topic analysis of the XP in Puyuma RTO and discuss its implications.

6.2 RTO as an embedded topic construction: a crosslinguistic look

It has been shown in Section 5 that the XP in Puyuma RTO has the status of an embedded topic, which shares a number of similarities with hanging topics in root clauses. The parallel behaviors between XPs and hanging topics in Puyuma thus suggest that languages may utilize parallel strategies in forming matrix and embedded topicalization. One question that arises from this observation is whether an embedded topic analysis may apply to other apparent cases of RTO.

It is widely observed that topics behave differently across and within languages. A topic may be base-generated in its theta-position and undergo A’-movement to the left periphery, commonly referred to as an internal topic; it may also be base-generated outside of a clause, as an external or hanging topic (e.g., Rizzi 1986; 1997; Cinque 1990; Aissen 1992; Anagnostopoulou 1997; Zeller 2009). If the XP in some RTO constructions has the status of an embedded topic, we expect to observe similar variation in behavior with the XPs.

This prediction is indeed borne out, along with Massam’s (1985) proposal that some ECM elements may have the status of an embedded topic. Besides Puyuma, Tsez and Romanian have also been analyzed as having an RTO construction which contains an XP as an embedded topic (Polinsky & Potsdam 2001; Alboiu & Hill 2013). Crucially, each of these three constructions exemplifies a different type of embedded topic construction: Puyuma RTO involves an XP base-generated as a hanging topic (80a), whereas the XPs in Tsez RTO have been shown to undergo covert A’-movement from the embedded theta-position to the embedded left edge (80b) (Polinsky & Potsdam 2001). Finally, in the Romanian RTO, it has been claimed that the XPs undergo cyclic movement from the embedded theta-position to a matrix A-position, as in (80c) (Alboiu & Hill 2013):

The differences in behavior of the XP in these three languages (80a)–(c) has an important implication, namely that the variation observed in topics across languages is also attested with XPs in apparent cases of RTO constructions. This strongly suggests that at least a subclass of RTO constructions may be properly analyzed as instances of embedded topicalization, calling for further investigation of existing cases of RTO and their correlation with topicalization in the same language.