Ā-Probing for the Closest DP

A central concern of syntactic theory is how nonlocal dependencies are formed and constrained. Since Chomsky 2000, 2001, much of this work has been fruitfully discussed in terms of probes and their specifications. A probe initiates a search for a goal that matches a particular feature specification, to Agree with or to Move. This article contributes to our understanding of the possible feature specifications and behaviors of probes that trigger Ā-movement.

For example, we may describe wh-movement in a language like English as involving C probing for the closest [WH] constituent (see, e.g., Rizzi 1990). This allows for wh-movement of the embedded subject in (1a) or the embedded object in (1b); in either case, the moved goal is the closest constituent with a [WH] feature. Intervening non-[WH] constituents are ignored. When multiple potential goals are accessible to the probe, the structurally closest goal is chosen, as reflected in the contrast in (1c).

Concretely, we adopt the definition for closest in (2), which is equivalent to the locality condition on Agree stated in the discussion in Chomsky 2000:122.

There are, however, languages with much stricter restrictions on Ā-extraction, such that only particular types of arguments can be Ā-extracted. Aldridge (2004, 2008) develops one influential approach to the analysis of so-called syntactic ergativity, narrowly defined as a ban on the Ā-extraction of transitive subjects. (See Deal 2015, 2016 and Polinsky 2017 for overviews and discussion.) One component of Aldridge’s analysis is a claim that Ā-probing can be restricted as in (3).

Combined with a commonly adopted approach to the clause structure of a subtype of ergative languages, a probe of this type will necessarily target absolutive arguments. Ā-probing of the form in (3) has been adopted for the analysis of Philippine-type Austronesian languages, for which the analysis was developed, as well as for similar extraction restrictions in Mayan languages in more recent work (Levin 2018, Coon, Baier, and Levin 2021). However, probing of this form has not been commonly invoked in the analysis of nonergative languages, leading some linguists to the impression that probing of this form is a special property of ergative languages.

In this article, we argue that Ā-probing for the closest DP is indeed attested in the grammars of nonergative languages, manifest in extraction constructions with apparent subject-only restrictions. In particular, Ā-probing for the closest DP makes accurate predictions for apparent exceptions to subject-only extraction restrictions: for example, even in a language where the subject is frequently the highest DP in a clause, if the language has a strategy for raising a non-subject DP to a higher position, such raising may feed the restricted extraction. We conclude that Ā-probes indeed can be specified to necessarily target the closest DP, as Aldridge proposes, and that such Ā-probes are not limited to ergative languages.

After reviewing the motivation for this conjecture as part of the analysis of syntactic ergativity in section 2, we formalize this mode of probing and present two novel arguments for it from relativization in Turkish and Rejang in section 3. In section 4, we then review and highlight relevant results from Keenan and Comrie’s (1977) typology of relativization and related subsequent work, which serves to motivate and contextualize the research reported here. In section 5, we argue that this restriction on Ā-probing to the closest DP must be a specification on individual probes, rather than a language-level or construction-level parameter. We conclude in section 6.

We begin by reviewing Aldridge’s (2004, 2008) analysis for syntactic ergativity in “Philippine-type” Austronesian languages, such as Tagalog. Philippine-type languages are verb-initial, with case-marking patterns that can be analyzed as exhibiting ergative-absolutive alignment.¹ Among core arguments of the verb, these languages allow only for Ā-extraction of the absolutive DP.

Aldridge’s theory for this extraction restriction is one specific instantiation of what Deal (2016) calls the “standard theory of syntactic ergativity,” narrowly referring to the absolutive-only extraction restriction. The shared intuition of these proposals is that transitive objects (O) canonically occupy a structural position above that of transitive subjects (A).² For Aldridge, in a transitive clause with two core arguments, an agent and a theme, the theme will move to the outer specifier of vP.³ The agent is base-generated as the inner specifier of DP. See (5). The verb is ultimately pronounced higher, preceding its arguments.

Aldridge also discusses clauses with applicatives, where the DP moved to the outer specifier of vP is a goal, instrument, location, or possessor (see Nie 2019) instead of a theme.

We now turn to Ā-extraction. Ā-extraction of DPs in Tagalog is limited to the extraction of absolutive arguments (e.g., transitive objects (O) and intransitive subjects (S)); transitive subjects (A) cannot be Ā-extracted. Aldridge proposes that Ā-probing by C necessarily targets the closest DP.⁴ Following the proposed structure for transitive clauses, Ā-extraction from a transitive clause will thus necessarily target the outer specifier of vP, which may be a transitive object (O) or an applicativized argument (see above). See (6). There is no way to target a transitive subject (A) for Ā-movement.⁵

In an intransitive clause, the sole DP argument (S) is the closest DP to the probe and thus can be Ā-extracted as well.⁶ This derives the syntactically ergative Ā-extraction restriction.

Aldridge’s analysis for the extraction asymmetry in Philippine-type languages thus relies on the conjecture stated in (3), repeated here:

This conjecture has been adopted as part of the analysis of syntactically ergative extraction asymmetries in other languages as well, including the ones proposed by Levin (2018) and Coon, Baier, and Levin (2021) for a number of Mayan languages. Also see Erlewine and Lim to appear for an investigation of extraction asymmetries in Bikol, a sister language to Tagalog, which strengthens the empirical case for extraction asymmetries in Philippine-type Austronesian languages to be based on (7), over and above Aldridge’s original argumentation.

We should step back, however, and note that Ā-probing of this form—if it exists—is perhaps unusual and conceptually surprising.⁷ Such probing would lead to an Ā-extraction process that has the locality profile of A-movement, rather than the familiar long-distance and “relativized” character of Ā-movement (as in, e.g., Chomsky 1977, Rizzi 1990; see (1)). It also raises the questions of whether non-DPs can be Ā-extracted in such a language and if so, how; we return to these questions in section 5.3. We also note that alternative accounts for the extraction behavior of such Philippine-type languages exist, which do not require Ā-probing that is restricted to the closest DP (7).⁸ In the pursuit of a maximally restrictive theory of grammar, then, it is tempting to reject the possibility of probing of the form in (7), or perhaps to somehow limit its availability to ergative languages.

The goal of this article is to show that there in fact is substantial motivation in nonergative languages for the existence of Ā-probing that is restricted to the closest DP (7). This conclusion in turn lends support for the plausibility of Aldridge’s proposal for absolutive-only extraction restrictions in ergative languages as well.

Given the potentially unusual nature of the idea that an Ā-probe would be limited to attracting the closest DP goal (7), as discussed above, we seek independent motivation for this possibility in grammar from beyond its original application to syntactic ergativity. In this section, we present two case studies of relativization in two nonergative languages, Turkish and Rejang, which we claim are best analyzed as involving an Ā-probe that can only attract the structurally closest DP. This approach will allow us to account not only for the basic subject-only restriction on these Ā-processes, but also for its apparent exceptions.

Let us begin by detailing, in schematic terms, the behavior that we expect from Ā-probing for the closest DP (7).⁹ Here, we describe such probes as a particular version of a probe that seeks a goal that bears both an Ā-feature (e.g., [WH], [REL]) and an A-feature ([D]), which we notate [PROBE:Ā+D] in the general case. This probe is additionally required to match with a goal that is the closest DP to the probe, where closest is defined as in the discussion of Agree in Chomsky 2000:122, repeated here from (2):

With [PROBE:Ā+D] restricted to the closest DP, we expect extraction of DP₂ in (9) to be impossible: DP₁ c-commands DP₂ and therefore counts as structurally closer to the probe.

This logic makes two predictions. First, if a language has an independent mechanism for bringing DP₂ above DP₁, we expect the probe to then be able to interact with DP₂.

Second, if there is no c-command relationship between two DPs, we might expect either to be extractable, as neither is closer to the probe than the other. (See again the definition of closest in (8).) One such configuration would arise when DP₂ is at the edge of DP₁.

The possibility that two constituents in this configuration both count as closest for a higher probe is also discussed and invoked in Pesetsky and Torrego 2001:363.

We now present evidence from relativization in Turkish and Rejang that motivates the existence of Ā-probing for the closest DP. In particular, we will show that the configurations in (10) and (11) indeed allow for Ā-extraction of DP₂ in constructions that disallow extraction in the configuration in (9). Neither language is morphologically ergative. We also situate our proposal in relation to work (e.g., Van Urk and Richards 2015, Bossi and Diercks 2019, Colley and Privoznov 2020, Scott 2021) that argues for probes of the form [PROBE:Ā+D] yet allows those probes to skip closer partial matches.

Turkish has two forms for relative clauses, shown in (12), traditionally described as a subject/nonsubject distinction (e.g., Underhill 1972, Hankamer and Knecht 1976). We follow Cagri (2005, 2009) in glossing these forms ‘subject relative’ (SR) and ‘nonsubject relative’ (NSR). Here, we concentrate on the behavior of relativization with the SR suffix, which receives a simple analysis as the exponent of a head that has an Ā-probe specified to seek the closest DP. We will discuss the syntax of NSR forms below.

It is well-known that, under certain limited circumstances, nonsubjects may be extracted using the SR form (see, e.g., Underhill 1972, Hankamer and Knecht 1976, Kornfilt 1984, 1997, 2000, and subsequent work). For instance, object relativization with the SR form is possible in examples such as (13a), where the subject is indefinite and in a low, immediately preverbal position.¹⁰ The source for this relative clause in (13a) is therefore (13b), where the accusative object is the highest DP in the clause. In contrast, object relativization across a high, nonindefinite subject must use the NSR form, as in (12b).

Temporal relatives may also be constructed with the SR form if the subject is a low, indefinite subject, as in (14). This too is explained by our analysis for the SR probe, as temporal adjunct DPs naturally occupy a position above low indefinite subjects.

Such examples show that the apparent subject orientation of the SR form is really a requirement to extract the closest nominal, rather than to extract the subject per se. We propose that the SR form reflects the use of [PROBE:REL+D] restricted to extract the closest nominal. Movement of the object (DP₂ in (15)) across a low, indefinite subject (DP₁) allows for the SR probe to target the object.

Possessor relativization presents a further class of apparent exceptions to the subject orientation of SR relatives. Example (16a) shows that the possessor of a subject is extracted using the SR form. Note that subjects in NSR clauses as in (16b) receive genitive case, as seen in the grammatical (12b), whereas subjects of SR clauses as in (16a) retain their unmarked, nominative case.

This possibility of possessor relativization is also explained by our analysis of SR as involving Ā-probing for the closest DP and our definition of closest in (8). Both the subject (DP₁ in (17)) and its possessor (DP₂) count as closest for the probe, as the former does not c-command the latter.

Also as our approach predicts, SR relativization can target the possessor of an object if the subject is in the low, immediately preverbal position (18). The derivation for (18) simply combines movement of an object above an indefinite subject that stays low, illustrated in (15), with the SR probe targeting the possessor of the highest DP, which counts as closest, as in (17). This derivation is illustrated in (19).¹¹

In contrast, if the subject is higher than the object—whether a bare definite or a birkaç ‘some’ indefinite—relativization of the object’s possessor must take the NSR form.

The behavior of relativization with the SR form shows the clear hallmarks of Ā-probing for the closest DP. SR relativization often targets subjects, which are generally highest in the clause, but can also target nonsubjects when the subject is exceptionally low, as well as possessors of the highest DP, which also count as closest. This suggests that SR forms are regularly subject-oriented because the subject is regularly the highest DP in the clause, not because the SR form specifically reflects other subjecthood properties of the goal.¹²

Next, we address the structure of relatives with NSR morphology and their relation to relatives with SR morphology. The above data show that SR and NSR forms are in complementary distribution, with the SR form appearing specifically where relativization can use Ā-probing for the closest DP, and the NSR form appearing otherwise.¹³ We propose that this reflects a preference for the use of SR relativization where possible, which in turn reflects a more general and familiar pressure for grammar to minimize probe-goal relations while maximizing featural matching.

Specifically, we propose that in any relative clause, the complex probe on the relative complementizer, [PROBE:REL+D], probes first. If this joint probing finds a matching goal, that goal is extracted and the complementizer then realizes SR morphology. If it does not find a match, the head is able to probe separately as [PROBE:REL] and [PROBE:D]; the former process will attract a [REL] goal that was inaccessible using Ā-probing for the closest DP, and the latter will copy back subject ϕ-features. The resulting complementizer will then be spelled out as the -DIK suffix with subject ϕ-agreement, which is more generally the default form for nominalized finite embedded clauses in the language. Here, use of the complex probe is privileged over use of two simple probes, since a perfect match for the complex probe will check two features with only one probe-goal relationship.

Probe interactions very similar to this have been proposed by Van Urk and Richards (2015) for Dinka and by Bossi and Diercks (2019) for Kipsigis. In each case, there is a probe that seeks a goal using a combination of A- and Ā-features. If there is an accessible goal carrying both features, the probe targets it, due to a principle that Van Urk and Richards (2015) call Multitasking. But if this joint probing fails, the head instead probes separately for a goal with the A-feature and a goal with the Ā-feature.¹⁴ There are just two differences between these interactions in Dinka and Kipsigis on the one hand and Turkish on the other. First, in Dinka and Kipsigis probing separately results in multiple movement, whereas in Turkish it results in movement of the [REL] goal and ϕ-agreement with the DP. Second, in Turkish probing jointly for a goal with both A- and Ā-features requires Ā-probing for the closest DP, whereas in Dinka and Kipsigis closer partial matches can be skipped.¹⁵ Abstracting away from these two differences, however, the mechanism that yields apparent “competition” between SR and NSR forms is equivalent to that proposed in these prior works.¹⁶

Finally, we note that the ability of an unsatisfied complex probe such as [PROBE:Ā+D] to subsequently probe separately for Ā- and D-goals must be subject to lexical variation, rather than being a general property of complex probes. In particular, in the case studies we discuss below, as well as those in section 5, we show that if Ā-probing for the closest DP is unable to find a suitable goal, the relevant Ā-construction is simply impossible.

Rejang is an Austronesian language spoken in southwest Sumatra. Ā-extraction of DPs in the language is limited to the highest nominal, whether or not that is the subject.¹⁷

Rejang is canonically SVO. Transitive verbs have two forms, active and passive. Active themes and passive agents cannot be Ā-extracted (with one exception, discussed below). For example, transitive theme relativization requires first promoting the theme to subject position using a passive.

There is evidence that this apparent subject-only restriction on extraction in Rejang is in fact better described as a requirement that only the highest nominal may undergo extraction. Evidence for this comes from three sources: agent cliticization, long-distance extraction, and extraction of possessors, which we discuss in turn.

McGinn (1998:372ff.) notes one exception to the ban on active theme extraction: extraction of active themes is possible if the agent subject is a clitic pronoun, as in (22).¹⁸

As McGinn argues, the agents in such cases are not conventional preverbal subjects, but instead occupy a lower position and are cliticized to the verb. Consider the contrast between the baseline canonical declarative in (23a) and the grammatical active theme relative in (23b), both with the preverbal inchoative auxiliary mulaé and the active verb ‘watch’. The canonical subject position precedes preverbal tense/aspect auxiliaries (23a), but a pronominal agent may appear in immediately preverbal position (23b), which allows the active theme to be extracted. Note also that the first person singular pronoun uku appears in the reduced form ku when immediately preverbal.

As McGinn (1989) notes, the structure here is reminiscent of the so-called passive type 2 in nearby Malayic languages (see, e.g., Chung 1976, Arka and Manning 1998). This is illustrated with Indonesian in (24), where an agent of limited or reduced size (see Nomoto 2021) immediately precedes the lexical verb and then allows for theme extraction.

However, importantly, the Rejang structures (22) and (23b) retain active voice morphology on the verb, whereas no voice morphology appears in the Malayic structures exemplified by (24). See McGinn 1989 for further discussion and comparison of these forms in Rejang and Malayic.

Our proposal straightforwardly accounts for the possibility of active theme extraction with agent cliticization in Rejang, as in (22) and (23b). It is well-documented that clitic arguments do not intervene for extraction in the same way that full DPs do (McGinnis 1998, Anagnostopoulou 2003). Presumably, cliticization of an argument to the verb renders it invisible for higher [D]-sensitive probes; or it is skipped, as it is not eligible for movement (see also Branan 2022). As the subject is a clitic, the object is now the closest DP to the probe on C.

Next, we turn to long-distance relativization. First, we note that in nonextraction contexts, embedded clauses take the complementizer bawo, in contrast to gi in the extraction examples above. This alternation will be important for the discussion that follows.

Long-distance subject extraction in Rejang comes in two forms. In one, the embedded clause has the complementizer bawo, with a resumptive pronoun in its subject position (26a). In the other, the embedded clause is headed by gi, with a subject gap (26b). In both cases, the higher clause must appear in the passive.

We first consider these complementizer and gap/resumptive alternations. Both reflect strategies for getting the nominal target for relativization to the edge of the embedded CP. The nominal can be base-generated at the embedded clause edge and bind a local pronoun (26a) or it can move from embedded subject position using the complementizer gi (26b).

As proposed by McGinn (1998:368–369), the entire embedded CP itself is then moved to the higher subject position via passivization of ‘say’, although it is then pronounced to the right, leaving the CP gap indicated in (26).¹⁹ Movement of the embedded clause to this higher subject position causes the DP at the embedded clause edge to be the highest nominal in the clause, allowing the relative complementizer gi to extract it.

In addition, long-distance relativization may proceed across a higher active verb if the higher verb’s agent is a preverbal clitic pronoun; see (28a–b). Per the discussion above, clitic pronouns are skipped, making the highest DP in the embedded clause the closest DP for the higher Ā-probe. In (28a), this is the embedded subject. Example (28b) furthermore shows long-distance theme relativization across multiple clitic pronouns, each complementizer along the way attracting the closest DP, skipping an agent clitic.²⁰

Further evidence that the extraction restriction in Rejang reflects [PROBE:Ā+D] on C, and is not specifically limited to subjects, comes from possessor relatives. Possessor relativization in Rejang is possible with a resumptive pronoun, but generally only for possessors of subjects.

However, relativization of object possessors is possible if the subject is a clitic pronoun, as attested in Dibul’s (2019) translation of the Quran into Rejang.

These examples show that possessor relativization is locality-sensitive and does not reflect a free process of pronominal binding. The use of the gi complementizer—which correlates with movement, as illustrated in (26)—also supports the view that possessor relativization as in (29a) and (30) involves extraction.

We can imagine two possible analyses for this possessor relativization: one where the target of relativization originates in the possessor position, with its trace pronounced as the pronoun ne (as in a recent proposal in Jeoung 2018 and as suggested by a reviewer), or with the target of relativization being generated at the edge of the DP and locally binding the possessive pronoun. In either case, the restriction of possessor relativization to possessors of the highest nominal (normally the subject) is explained by our analysis as analogous to the long-distance extraction facts above: the target in these cases is at the edge of the highest DP and therefore counts as closest to the probe.

In sum, a close look at the apparent subject-only extraction restriction with the gi complementizer in Rejang—and the shape of its various exceptions—provides strong motivation for the theory of probing presented here. Rejang allows for the extraction of nonsubject nominals across clitic agents with gi, and it provides strategies to place nominals at the edge of subjects—both nominal and clausal—causing those nominals to become the closest nominals to the probe and therefore licit targets for extraction.

We now turn to the broader typology of extraction asymmetries for further evidence of the possibility of Ā-probing for the closest DP (7). Our discussion will center around Keenan and Comrie’s work on the typology of relativization (Keenan and Comrie 1977, 1979 (K&C), Comrie and Keenan 1979). In this work, K&C claim that there exist languages with relativization strategies that apply specifically to subjects but to no other types of arguments.²² As Deal (2015:698–699) notes, such forms of relativization may be candidates for being analyzed as involving Ā-probing for the closest DP.

Revisiting all the subject-only relativization strategies identified by K&C, we note that the level of detail they provide on individual languages is generally insufficient to determine whether any of these relativization strategies are best analyzed as involving Ā-probing for the closest DP. Thankfully, however, more detailed subsequent studies exist for some of these languages. We conclude that some but not all of K&C’s subject-only strategies provide further evidence for our conjecture. This section thus serves two purposes: First, it highlights a few more case studies—from Arabic, Māori, and Toba Batak—which we take to provide strong supporting evidence for the possibility that Ā-probes may be restricted to the closest DP. Second, it serves as a note of caution against taking K&C’s results to have already established the necessity of Ā-probing for the closest DP in grammar, thereby underscoring the motivation for this article.

In their survey, K&C report eleven languages with relativization strategies that can target subjects but not other nominals: Northeast Aoba/Ambae, Arabic, Iban (Sea Dayak), Javanese, Kera, Kiribati (Gilbertese), Malagasy, Māori, Minangkabau, Tagalog, and Toba Batak. Of these, only Malagasy and Toba Batak are discussed in any detail in K&C 1977. With the exception of Arabic and Kera (East Chadic), all of these languages are Austronesian.²³

We first discuss Kiribati (Oceanic; VOS), which K&C describe as utilizing a gap strategy for subject relatives (32a) but a pronoun strategy for object relatives (32b). (This same description applies to Northeast Ambae (also Oceanic), Arabic, and Kera; we discuss Arabic below.) Object pronouns appear on the verb with the linker -i-.²⁴

At first glance, such a language seems amenable to an analysis utilizing an Ā-probe that targets the closest DP: Ā-extraction is limited to the structurally highest DP, the subject, and relativization of other arguments necessitates the use of resumptive pronouns.

However, this view is challenged by data on long-distance object extraction in Sabel 2013. In (33), fronting the embedded plural object ‘Mary and Tien’ triggers a third person plural pronoun on the embedded verb ‘love’, as expected, but also on the higher verb ‘know’.

On the basis of such examples, Sabel argues that the object “pronoun” on Kiribati object extraction verbs is a form of agreement fed by successive-cyclic movement of the object. Thus, the difference between subject and object relativization illustrated in (32) may simply be morphological. Kiribati thus represents an instructive test case where further work on the language casts doubt on taking K&C’s reported behavior as a natural candidate for an analysis using Ā-probing for the closest DP.²⁵

K&C’s description of relativization as subject-only in Malagasy and Tagalog also does not immediately necessitate Ā-probing for the closest DP. Both are Philippine-type Austronesian languages explicitly discussed by Aldridge (2004, 2008) as amenable to her analysis for syntactic ergativity reviewed above. However, alternative approaches for these languages exist that do not involve such an Ā-probing restriction; see ^{footnote 8}.

In some cases, however, subsequent work on the subject-only relativization strategies identified by K&C has led to forceful arguments for Ā-probing that is limited to the closest DP. Such is the case for Toba Batak (Erlewine 2018), where relativization is limited to the subject (active agent or passive theme), relativization of transitive themes being fed by passivization. (K&C also give this same description for Iban, Javanese, and Minangkabau—all Indonesian-type Austronesian languages (see ^{footnote 1}).) Although Cole and Hermon (2008) have proposed that this restriction reflects the “frozen” nature of the nonsubject DP arguments, Erlewine (2018) shows that, under certain circumstances, a subject DP and a nonsubject DP can be fronted simultaneously. Erlewine argues that the basic extraction restriction thus must be due to Ā-probing being limited to the closest DP—with options for further probing leading to multiple extractions—rather than being due to the general immobility of nonsubject DPs as Cole and Hermon propose.

Further work on Māori has also led to an analysis involving Ā-probing for the closest DP. Douglas (2018) discusses the fact that clefting in Māori—built from a kind of headless relative clause—can only target subjects of verbal and prepositional phrase predicates, not subjects of nominal predicates. He argues that this is because the Māori cleft construction uses Ā-probing for the closest DP; in this predicate-initial language, a nominal predicate is higher than its argument. However, as with their description of Toba Batak, K&C’s initial description of Māori as having a subject-only relativization strategy does not by itself force a limited Ā-probing analysis; only with further investigation can a decisive argument for restricted probing be made.

Finally, we discuss the analysis of relativization in Arabic due to Shlonsky (1992). Arabic allows extraction of subjects but requires pronominal resumption for extraction of all other arguments. Shlonsky proposes that Spec,CP in Palestinian Arabic is an A-position, rather than an Ā-position, and thereby obeys the locality profile of A-movement. Although not cast in the contemporary probe-goal terms used here, Shlonsky’s analysis amounts to proposing that the Ā-probe for relativization must target the closest DP. Similar proposals have since been advanced for new-information clefts in French (Belletti 2015) and topicalization in southern Bantu (Bliss and Storoshenko 2009, Pietraszko 2021), which target subjects.²⁶

In sum, although the evidence provided by K&C (1977, 1979; see also Comrie and Keenan 1979) is by itself generally insufficient to motivate the existence of Ā-probing restricted to the closest nominal, stronger arguments for proposals of this form have been subsequently developed for three of the languages in their study—Arabic, Māori, and Toba Batak—all of which are, again, clearly not ergative. These three case studies thus join our discussion of Turkish and Rejang to form a compelling reason to think of Ā-probing for the closest DP as a true possibility in grammar, and not only in ergative languages. This evidence in turn indirectly supports the feasibility of Aldridge’s (2004, 2008) approach to syntactic ergativity presented in section 2.

We have now established that Ā-probing for the closest DP is indeed a strategy employed by the grammar of nonergative languages. At the same time, we know that languages also employ relativized Ā-probing that can skip intervening nominals without the matching Ā-feature; see (1). In this section, we turn to the nature of this variation. We show that Ā-probing for the closest DP is neither a language-level parameter nor a construction-level parameter. Instead, we argue that this choice of restricted probing is made with respect to individual heads.

We begin by discussing relativization in Haya, a Bantu language of the Great Lakes region of Africa. Haya demonstrates that heads at different positions in a single Ā-construction can differ in their choice to employ Ā-probing for the closest DP or not.

Local relativization in Haya can target both subjects and objects.²⁷

Duranti (1977) shows that, in long-distance relativization of an object in Haya, the object must be promoted to subject before undergoing further Ā-movement to the final landing site in the matrix clause, as demonstrated through the contrast in (35b–c).

We can understand this effect as resulting from intermediate movement being driven by [PROBE:REL+D] on embedded C that must target the closest DP, even though the highest clause of the relative is not restricted in this way.²⁸ Promotion of an embedded object to subject position through passivization causes the theme to be the highest nominal in the embedded clause, allowing it to then be probed from above and thereby extracted.

For the analysis developed here, whether or not the extraction restriction obtains in a configuration involving C is a function of the lexical items in the context. It is relatively common for embedded complementizers to differ from matrix complementizers in terms of what they attract; consider, for instance, the presence vs. absence of T-to-C movement and do-support in standard English matrix and embedded questions. Haya, then, is a language in which the outermost relative complementizer (abo in (35b–c)) simply probes for [REL], but the embedded complementizer ati must find a [REL] target that is the closest DP. Only extraction out of clauses headed by ati exhibits the subject-only restriction, even within a single long-distance extraction chain.

Late Archaic Chinese (LAC) exhibits a number of extraction asymmetries that are attributable to restricted Ā-probing by [PROBE:Ā+D] as described here. At the CP level, Aldridge (2019) shows that only subjects can be Ā-extracted to the clause edge in LAC and therefore proposes that these processes involve Ā-probing for the closest DP. In the interest of space, we will not review this evidence from movement to Spec,CP here. Instead, we call attention to a restriction on the behavior of nonsubject wh-phrases that undergo Ā-movement to a clause-medial position in LAC (Aldridge 2010).

While the canonical word order of LAC is SVO, Aldridge (2010) shows that wh-objects in LAC appear preverbally.²⁹

As Aldridge (2010, 2019) notes, basic ditransitives in LAC take their theme as their second object, in “V recipient theme” order. With certain verbs, there is also an option to bring the theme in front of the verb using the object marker yǐ. Aldridge (2010) observes that only the first object of either of these ditransitive predicates can move; in particular, a theme wh-word in second object position is left in situ as in (37a), whereas a theme wh-word that precedes the other internal argument is then fronted as in (37b).

These facts support the view that Ā-probing for the closest DP is an option for movement-driving heads in general, and not just with C. In LAC, the head that triggers clause-medial movement (perhaps v, following Aldridge 2010, 2019) bears [PROBE:WH+D] restricted to seeking the closest DP.³⁰

The existence of Ā-extraction constructions that must target the closest DP naturally leads to the question of whether and how non-DP constituents can be Ā-extracted. As we have argued, Ā-probing for the closest DP is not a language-level parameter. A language that utilizes [PROBE:Ā+D] in some configurations may also Ā-move non-DPs, but we predict that this will involve different heads or probes.

Here, we return to Tagalog—one of the languages that motivated the idea of Ā-probing for the closest DP in Aldridge’s work—and show that the extraction of DPs and the extraction of non-DPs indeed behave differently, exemplified here with wh-questions. DP-fronting as in (38a) must target the absolutive argument—here, the object of ‘play’—which appears with an ang case marker in preverbal position. Fronting of a non-DP as in (38b) lacks this ang marker. In addition, note that the second-position clitic pronoun =nila encliticizes to the verb in (38a) but to the wh-phrase itself in (38b). See Hsieh 2020a,b for in-depth work on these non-DP extraction constructions in Tagalog.

Structural differences are also observed between DP and non-DP extraction in Rejang. Recall that relativization over DP arguments must target the closest DP. Such examples allow for an optional relative pronoun before the complementizer, as in (39).³¹

Relativization in Rejang can also target prepositional objects. In such cases, a wh-containing PP fronts to a position following the complementizer.

We can understand prepositional object relatives as in (40) as involving a separate, optional process of PP-fronting described by McGinn (1982:10); this movement “smuggles” the relative operator to a position above the subject for successful targeting by [PROBE:Ā+D]. Although these relative operators are normally optionally realized as a wh-word in Spec,CP (as in (39)), the morphological needs of the preposition require that it be pronounced overtly next to the preposition in (40).

Such structural differences between DP and non-DP extraction are unsurprising under our account. As Aldridge 2004 and much subsequent work has argued for Tagalog, and as we have argued for Rejang, DP-movement takes place via an Ā-probe that must target the closest DP, [PROBE:Ā+D]. As this probe cannot target non-DPs, a separate, second Ā-probe—for instance, simply [PROBE:Ā], potentially on a different functional head—must be involved in the fronting of non-DPs.

An important remaining question is why these separate Ā-probes cannot also target DPs, potentially leading to the extraction of nonclosest DPs. One possibility, suggested by Erlewine (2018) and Hsieh (2020a) for Toba Batak and Tagalog, respectively, is that the Ā-probe that leads to successful movement of non-DPs cannot target nonclosest DPs because doing so would bleed case licensing on the fronted DP. For more details, see these accounts and the argumentation there.

In this article, we argued that Ā-probes can be restricted to target only the closest nominal, as originally proposed by Aldridge (2004) as part of an analysis for syntactic ergativity. We showed that this (possibly conceptually surprising) form of restricted Ā-probing is well-attested in nonergative languages. Probing of this form in many languages gives rise to what at first glance may appear to be a subject-only extraction restriction. However, on closer examination certain nonsubjects can also be be extracted via processes that rearrange nominals as well as via subextraction from highest nominals. We illustrated this type of extraction restriction in Turkish and Rejang in detail and provided references to other work describing such facts in Arabic, Māori, and Toba Batak—all nonergative languages.

Establishing the existence of Ā-probing restricted to the closest DP contributes to the growing literature on possible interactions between A- and Ā-features in probe specifications (e.g., Van Urk 2015, Van Urk and Richards 2015, Baier 2018, Erlewine 2018, Bossi and Diercks 2019, Colley and Privoznov 2020, Coon, Baier, and Levin 2021, Scott 2021) and on the space of variation therein. We furthermore clarified that Ā-probing for the closest DP is a property of specific probes on heads, rather than a language- or construction-level parameter. In particular, languages with constructions that involve Ā-probing for the closest DP often also have other strategies for Ā-extraction that are not so restricted.³² What this implies, then, is that these grammars must also have mechanisms for choosing between different extraction strategies. We pointed to two such mechanisms here: the ability of some complex probes to prefer full matches but optionally allow partial matches (see discussion at the end of section 3.1) and the ability of certain movements to bleed case licensing and therefore be unable to target DPs (section 5.3).

For discussion of specific facts reported in the article, we thank Henrison Hsieh (Tagalog) and Jaklin Kornfilt (Turkish). For additional discussion related to other languages in the Keenan and Comrie survey that are unfortunately not discussed here for reasons of space, we thank Dan Brodkin (Minangkabau), Catriona Hyslop Malau (Northeast Aoban/Ambae), and Alex Smith and Carly Sommerlot (Iban). We also thank Amy Rose Deal and Asia Pietraszko for related discussion, and Justin Adhiyatma for preparing our Rejang corpus materials. This research is supported by the Singapore Ministry of Education under grant MOE2017-T2-2-094 “Subjecthood in Southeast Asia: Description and Theory” to the second author.