Ch’ol is a morphologically ergative language in which grammatical relations are head-marked via two sets of morphemes: Set A (ergative, possessive) and Set B (absolutive); all core arguments may be pro-dropped. As in other Mayan languages, predicate stems can be divided into “verbal” (generally eventive) and “non-verbal” (generally stative) classes. Verbal stems typically appear with a “status suffix” which varies based on transitivity, and require one of three aspects: perfective (tyi, tsa’, ta’), as in (6a); imperfective (mi, muk’, mu’); or progressive (choñkol, woli). Non-verbal predicates, as in (6b), may not appear with aspectual inflection and do not have status suffixes.

Basic constituent order is VOS/VS, with clause-initial topic and focus positions (Vázquez Álvarez 2011; Clemens & Coon 2018). Vázquez Álvarez (2011) argues that Ch’ol topics are external topics in the sense of Aissen (1992), preceding the clause-internal focus position, as in (7).

Focussed constituents, wh-elements, relativized nominals, and other focus-sensitive operators all compete for the single position clause internal left-periphery position (“Focus” in (7); Norman 1977; Aissen 1992; Velleman 2014). Wh-movement is obligatory for question formation in Ch’ol and multiple wh-questions or wh-in situ are impossible in the language; see Vázquez Álvarez & Coon (2020) for examples and further details. While some Mayan languages restrict the extraction of transitive subjects, known in the Mayanist literature as the Ergative Extraction Constraint (Aissen 2017b) (an instance of “syntactic ergativity”), Ch’ol does not. All core arguments freely extract (Vázquez Álvarez 2011).

Finally, we note that while clausal pied-piping has not been described in Mayan (though see §4 for a comparison with “inversion” in Popti’), pied-piping with inversion has long been documented in the nominal domain for wh-possessors (see e.g., Aissen 1996; Coon 2009; Little 2020; Aissen & Polian 2024), as well as with the wh-complements of relational nouns and prepositional phrases (Aissen 1996; Ewing 2022). Ch’ol permits both possessive phrase and PP pied-piping; see Coon (2009) and Little (2020) for wh-possessors, and Cable (2007: ch. 5) on PPs.

Vázquez Álvarez (2013) describes three types of embedded complements in Ch’ol: (i) non-finite; (ii) “less-finite”; and (iii) fully finite. Both non-finite and less-finite complements obligatorily lack aspect marking (taken in Aissen 1992 and subsequent work to occupy finite Infl⁰), are dependent on the matrix clause for their temporal interpretation, and exhibit obligatory control of their subjects by a matrix argument. Non-finite clauses, like the one in (8a), are obligatorily intransitive (including examples of transitive roots with incorporated objects), while less-finite clauses, like the one in (8b) are transitive. While the “less finite” form requires Set A subject marking on the embedded clause, it is obligatorily identical to the subject marking in the matrix clause. We take these clauses to be smaller than IP, but remain agnostic about their exact structure; following Aissen’s (2017a) discussion of complementation in Mayan more generally, below we refer to these two clause types together as “non-finite”.

Finite embedded clauses, in contrast, are distinguished from both types of non-finite clauses in (8) by: (i) the obligatory appearance of an aspectual marker; (ii) the possibility of a complementizer (optional for some speakers, preferred for others); and (iii) the absence of any dependency between matrix and embedded arguments (see Aissen 2017a). These properties are shown in (9).

The relationship between complement type and a control interpretation between matrix and embedded arguments can be seen clearly with the verb om ‘want’, which may combine with either non-finite or finite complements. As shown in (10), coreference with the matrix subject is obligatory in the non-finite complements in (10a). For disjoint reference between matrix and embedded subjects, a finite embedded clause is required, as in (10b).

The examples in (10) illustrate an additional contrast: while the aspectless non-finite complements appear in typical object position (recall that word order is VOS), finite clauses are obligatorily extraposed, appearing after the matrix subject (Vázquez Álvarez 2011). We remain agnostic about the nature of this extraposition, but take this to further motivate a difference in syntactic clause size between non-finite clauses, on the one hand, and fully finite embedded clauses on the other.

Only fully finite embedded clauses permit clausal pied-piping, the focus of the remainder of this paper. As shown in (11), for example, a wh-object of a non-finite complement must front to initial position; pied-piping of the dependent verb form is ungrammatical. We connect the impossibility of clausal pied-piping of non-finite clauses to the absence of higher clausal structure (i.e., the absence of a CP layer), which will form part of our analysis of clausal pied-piping below.

As noted in Vázquez Álvarez 2011, verbs which take finite complements in Ch’ol encode meanings typical of verbs which take clausal complements cross-linguistically (Cristofaro 2003; Noonan 2007), including utterance verbs like al ‘say, believe’, su’b ‘say, accuse’, jak’ ‘respond, admit’ (12); propositional attitude verbs like u’biñ ‘think, believe’, ch’ujbiñ ‘believe, accept, obey’, pi’tyañ ‘await, anticipate’ (13); verbs of knowledge such as ujil ‘know’, ña’tyañ ‘know, undertand’ (14); desideratives like om ‘want’, mulañ ‘like’ pi’tyañ ‘wait’, (15), and verbs of perception like ilañ ‘see’, k’el ‘see, watch’, chäñtyañ ‘watch, observe’, ñächtyañ ‘listen’ (16). All of the bracketed forms below can serve as matrix clauses minus the complementizer che’.3

We focus here on four of the verbs from section 2.2, shown in the table in (17). Two of these verbs, al ‘say, tell’ and mulañ ‘like’, allow for either realis or irrealis complements, while the other two, ilañ ‘see’ and om ‘want’, combine with only realis and irrealis complements, respectively.

We begin with the unambiguous forms, ilañ ‘see’ and om ‘want’. In declarative contexts, both combine with complement clauses headed by the complementizer che’, and both appear with aspect and person marking.

The examples in (20) illustrate that long-distance extraction of all core embedded argument types—intransitive subject, transitive object, and transitive subject—is possible from the complement of om. The complementizer che’ is consistently judged as either ungrammatical or dispreferred on the embedded clause any time it is crossed over by a wh-word; we illustrate this explicitly in (20), though it holds consistently for other long-distance extraction examples below.

Clausal pied-piping is starkly ungrammatical with the complement of om, shown in (21).

Turning to realis complements of ilañ ‘see’, we find a different pattern. Extraction of all core arguments triggers pied-piping with inversion of the wh-word, shown in (22). As with long-distance extraction, the complementizer che’ may not appear anywhere in these examples.

All speakers consulted find both pied-piping and long-distance extraction of the wh-word—shown in (23)—grammatical. Two consultants expressed a preference for the pied-piping option in certain cases, but consistently agreed that both possibilities were good. We were unable to discern a pattern in the preference for one choice over another, and further work is needed to determine what factors may determine the choice between the two.

Despite the fact that the overt complementizer che’ is impossible in both types of long-distance wh-dependency, these clauses nonetheless show no signs of being reduced. In addition to aspect marking, negation is also possible, in both pied-piped clauses like (24a), and clauses from which long-distance extraction has occurred, like (24b).

To capture the obligatory absence of che’, we assume that complementizers which drive wh-movement (both interrogative Cs, as well as embedded Cs responsible for “secondary wh-movement”, discussed in §4) are morphologically null.4

Having established the initial generalization, we now look deeper into the properties of realis and irrealis complement clauses, which will allow us to probe differences in the two verbs which take either realis or irrealis complements. First, however, we note that there is no obvious difference in syntactic size between the two embedded complement types. Vázquez Álvarez (2011) notes, for example, that fully finite embedded clause can be distinguished from the non-finite embedded clauses examined in section 2.2 by their ability to host fronted focussed constituents, negation, and second-position clitics. The sentences in (25) and (24) show that these properties hold not just for the realis clauses examined in Vázquez Álvarez (2011), but also for irrealis complements.

Given the appearance of the complementizer che’, a focus position, and obligatory aspect marking, we take these complements to be CPs.

Despite apparent similarity of syntactic size, there are two differences which distinguish between realis and irrealis embedded clauses. First, Ch’ol has a number of second-position clitics, among them the irrealis clitic =ik and the realis clitic =ta’ (Vázquez Álvarez 2011). The irrealis clitic may optionally appear on complements to desiderative verbs which take irrealis complements, like om ‘want’ (27), but is impossible in complements of realis verbs like il ‘see’ (28); =ta’ has the opposite distribution in embedded clauses.

Second, while it is the case that non-control/finite complements to om ‘want’ must occur with an aspect marker (as with other finite clauses), only the imperfective aspect is possible, as in (29a). Any other choice of aspect is ungrammatical, as shown with the ungrammatical progressive and perfective forms in (29b) and (29c).

These differences between realis and irrealis clauses will help us disentangle the realis and irrealis uses of the other two embedding verbs we focus on here, al ‘say, tell’ and mulañ ‘like’. As shown in (30), matrix al can embed realis or irrealis clauses, which may be morphologically identical. As expected, the embedded irrealis form in (30a) is necessarily in the imperfective aspect, while the aspect of the embedded realis form may vary with the context (i.e., whether he bought tomatoes yesterday, is currently buying tomatoes, etc.).

The strategy chosen for wh-extraction disambiguates. In irrealis contexts, long-distance extraction is the only possibility (31a); clausal pied-piping enforces a realis interpretation, (31b).5

While the sentence in (31b) is infelicitous in the context in (31), it is the preferred response to the question in the realis context in (32). As above, for some speakers the pied-piping option in (32a) is preferred; for others, either pied-piping or long-distance extraction is possible, (32b).

The verb mulañ ‘like’ similarly may take either irrealis or realis complements, as shown in (33). The irrealis complement in (33a) is necessarily imperfective and may appear with the irrealis clitic. The realis form in (33b), in contrast, is not aspectually restricted and =ik is impossible.

Extraction patterns follow what we saw above. Long-distance extraction in (34a) either favours or forces an irrealis interpretation; clausal pied-piping in (34b) only receives a realis interpretation.6

The above examples have all involved the extraction of argument wh-words, but the same patterns are found with adjunct wh-questions, like those in (35). Like its English equivalent, the question in (35a) is ambiguous: baki ‘where’ could be asking either about the time of you saying, or about the time of Ana going. The addition of the complementizer che’ in (35b) forces a matrix association, compatible with the generalization that clauses from which extraction has taken place may not have overt che’. Finally, in (35c) baki in the embedded clause triggers clausal pied-piping, and as expected, baki can only be interpreted as modifying the embedded verb. The possibility of the realis clitic =ta’ confirms that this pied-piped clause is realis.

Unlike certain inversion constructions in English and Popti’, clausal pied-piping in Ch’ol necessarily involves wh-movement (in particular, wh-movement, seen throughout, or focus movement, discussed below). Ross (1973) dubs constructions like the English in (36a) “slifting” (“sentence lifting”), proposing that the initial clause originates as a finite embedded clause; a transformational rule deletes the complementizer and moves the embedded clause to adjoin to the left periphery of the matrix clause. These bear some obvious resemblance to the Ch’ol clausal pied-piping examples above. More recently, Haddican et al. (2014) and Stepanov & Stateva (2016) discuss examples of “interrogative slifting”, as in (36b), noting the parallels to clausal pied-piping in Basque.

In a potentially similar vein, Craig (1977) describes certain “inversion” constructions in the Mayan language Popti’, which at first glance also look similar to the Ch’ol clausal pied-piping constructions introduced above. In particular, Craig notes that “inversion is preferred when an NP is questioned out of the complement sentence”, as shown by the pair of examples in (37), mirroring the Ch’ol preference for clausal pied-piping in long-distance extraction.

In Popti’, however, as in English slifting, inversion is not restricted to wh-extraction environments, shown in (38). Furthermore, inversion only occurs with three embedding verbs, all involving the root al ‘say’ (an apparent cognate with the Ch’ol root al ‘say, tell’): hala ‘to say’, ay -ala ‘to desire’, and ham -alni ‘to think’. Additionally, Craig notes that the inverted clause must be in a reduced aspectless form.

In contrast with English slifting and Popti’ inversion, clausal pied-piping in Ch’ol requires wh-movement within the embedded clause. Compare the grammatical example in (39a) with the ungrammatical (39b). Here we observe that attempting to move the non-interrogative clause results in ungrammaticality, regardless of the presence or absence of a complementizer.7

Given that wh-movement is obligatory in Ch’ol wh-questions (see §2.1), it is perhaps unsurprising that wh-in situ is ungrammatical within the pied-piped clause. The examples in (40)—involving a pied-piped clause with an in situ wh-word—show this to be true for object and subject wh-questions within the pied-piped clauses, respectively.

As an instance of wh-movement, we expect clausal pied-piping to be able to cross multiple clauses; this is indeed possible, and an important focus of section 4.3 below. Like other instances of wh-movement in Ch’ol, clausal pied-piping is also subject to island effects, shown for extraction out of an interrogative complement (i.e., a whether-island) in (41). The examples in (41b)–(41c) remain ungrammatical regardless of the placement of mi ‘if’.

While space prevents us from a detailed look at these effects, we note that in Basque, clausal pied-piping has been shown to repair some, but not all, island violations (Ortiz de Urbina 1989). We save a detailed comparison and understanding of this apparent difference for future work.

Finally, we note that while clausal pied-piping cannot cross an island, it can target the edge of an embedded interrogative CP, as in (42).

In a general survey of pied-piping, Heck (2008: 105) describes clausal pied-piping as a “rather rare phenomenon.” Nonetheless, clausal pied-piping has been described in a range of unrelated and typologically diverse languages—including Wolof (Niger-Congo; Torrence 2013), Tlingit (Na-Dene; Cable 2007), Imbabura Quechua (Cole 1982; Hermon 1985), and Basque (Ortiz de Urbina 1989; 1993; Arregi 2003)–where it has been argued to provide evidence for the successive cyclic nature of movement in long-distance wh-dependencies. Common themes include (i) the dependence of clausal pied-piping on wh-movement internal to the pied-piped clause—what Heck terms “secondary wh-movement”— and (ii) the absence of a semantic distinction between pied-piped and long-distance extraction alternations (Arregi 2003; Cable 2007; Torrence 2013).

This latter property distinguishes clausal pied-piping from wh-scope marking constructions in languages like Hindi and German, a focus of Arregi (2003). Specifically, contra suggestions in Horvath (1997) and Lahiri (2002), Arregi (2003) argues that clausal pied-piping in Basque has no interpretive consequences, setting it apart from wh-scope marking constructions in languages like Hindi, Hungarian, and German. Arregi (2003) demonstrates specifically that Basque clausal pied-piping involves only a single presupposition, on par with typical long-distance wh-movement constructions, and unlike wh-scope marking constructions. Ch’ol behaves identically to Basque, as shown by the examples in (43), modelled after equivalent Basque examples in Arregi (2003). Specifically, either the clausal pied-piping question in (43a) or the long-distance extraction question in (43b) is a natural question given the context.

If clausal pied-piping mirrored the behaviour of wh-scope marking, we would expect the pied-piped construction in (43a) to introduce two presuppositions: (i) that Juan killed someone, and (ii) that Maria thinks that Juan killed someone. Following Arregi (2003), given that (43a) is appropriate in a context in which it is clear that Juan hasn’t killed anyone, we confirm that Ch’ol behaves like clausal pied-piping in Basque—and unlike wh-scope marking constructions in Hindi and German.

In the sections above we have seen that clausal pied-piping in Ch’ol is: (i) possible only with realis CPs; (ii) that it is dependent on wh-movement within the pied-piped clause; and (iii) that it has apparently no semantic consequences, alternating freely with long-distance wh-extraction. A basic syntactic structure for clausal pied-piping in Ch’ol is provided in (44), following analyses of Basque (Ortiz de Urbina 1989; 1993; Arregi 2003) and Tlingit (Cable 2007) (setting aside QP for now to focus on the basic structure). Specifically, we find movement of the embedded wh-word to the edge of the embedded Spec,CP (CP₂). This may be followed either by continued movement of the wh-word alone to the specifier of the matrix clause, resulting in familiar long-distance extraction (not shown here), or by movement of the entire embedded CP to the specifier of the matrix clause, Spec,CP₁, as in (44).

The next question to address is how it is possible for CP₂ to be targeted for extraction, rather than the wh-word itself—the standard issue for pied-piping constructions. We assume that wh-movement is triggered by unvalued [uQ] features on interrogative C. One option for ensuring that CP₂ is targeted rather than the wh-word in its specifier is feature percolation, proposed for example by Ortiz de Urbina (1993) for Basque. Under a percolation account, the wh-word bears the valued [Q] features which ultimately drive wh-movement. To account for clausal pied-piping, the wh-word moves to the specifier of the embedded clause, and then its [Q] feature “percolates” to the phrasal projection which immediately dominates it, CP₂. Now that CP₂ bears the [Q] feature, it is targeted for movement to the specifier of the CP bearing [uQ]. This percolation operation would be optional in Basque and Ch’ol, accounting for the fact that either the wh-word or the embedded clause may be targeted for movement to Spec,CP₁.

An alternative account of pied-piping is proposed by Cable (2007), foreshadowed above, who argues that wh-movement is always mediated by a projection, QP, headed by a focus-sensitive operator, Q. Q is realized as an overt question particle in a language like Tlingit, but covert in other languages, like Ch’ol. QP ultimately raises to check the unvalued [uQ] features on the interrogative C. Optionality in where the Q head merges results in variation in pied-piping. In the long-distance extraction of a wh-word, the Q head merges directly with the wh-DP. To account for clausal pied-piping, Cable (2007) proposes that the Q head may instead merge with the embedded CP. Following Cable, secondary movement of the wh-word within the embedded clause is motivated by the need for the embedded wh-word to enter into a local relationship with Q; we assume that secondary wh-movement is driven by an [EPP] feature on the relevant embedded CP which triggers movement of the embedded wh-word to Spec,CP, ensuring this locality (see also Aissen & Polian 2024 on pied-piping with inversion in the DP domain). As argued in detail in Cable (2007), the choice of whether to merge Q directly with the wh-word, or with a higher projection containing the wh-word, does not have semantic consequences, so long as certain locality conditions are met.8

While a simple case of extraction does not provide an immediate way to distinguish between these two possible analyses, an example with multiple embedded clauses, like the sentence in (45), does; see Coon (2009) for analogous discussion in the case of DP-internal pied-piping in Ch’ol. In particular, we examine predictions for extraction of the embedded object bu’ul to the matrix clause, in order to form the question equivalent to ‘What did you say that Carol liked that Nico planted?’.

Both percolation and QP analyses correctly predict the possibility of extraction of the embedded wh-word alone to the matrix clause, as shown in (46). In a feature percolation account, no feature percolation takes place and the wh-word is directly targeted for movement through the edges of each CP. In a QP account, on the other hand, the Q head (null in Ch’ol) merges directly with the wh-DP, here chuki ‘what’. QP is then attracted to through the edge of each clause to form the matrix question. (As our aim here is to compare the two analyses, we do not represent either percolation or a QP projection in the examples below, but summarize the required mechanisms for each derivation in (50) below.)

Under a QP account of pied-piping, variation in pied-piping can be connected to the selectional properties of Q. In Ch’ol (and Tlingit, and Basque), Q must be able to select CP. A QP analysis thus predicts two additional options for the original sentence in (46): Q could merge with the most embedded CP₃, or Q could merge with the intermediate CP₂. The first possibility is shown in (47). Here, the wh-word chuki first undergoes “secondary” movement to the edge of CP₃; as noted above, we assume that secondary movement is driven by edge features, which ultimately ensure that the wh-word enters into local relationship with Q (indicated with ➀; see Cable 2007 for details). Next, CP₃, which is dominated by QP, is attracted to the edge of CP₂, and then further extracted to the edge of the matrix clause.

A feature percolation account—in which the relevant features begin on the wh-word itself—requires an analysis in which the wh-word first moves to the edge of CP₃ and its [Q] features percolate. Next, [Q]-bearing CP₃ moves to the edge of CP₂ and no percolation must occur; finally, CP₃ moves to the matrix clause.

For a QP account, the third predicted structure, not seen so far, is shown in (48). Here, the Q head selects the intermediate CP₂. The wh-word chuki undergoes secondary movement to the edge of the embedded CP₃, and then further to the edge of CP₂.9 Next, CP₂, dominated by QP, is attracted to the specifier of of the matrix CP. Though some Ch’ol speakers found this construction more difficult than the options in (46) and (47), indicated here with “?/✓”, it was judged to be acceptable (and in clear contrast with the next example discussed below).

Under a feature percolation account of the clause in (48), the wh-word would hold its features after movement to Spec,CP₃, and then [Q] features would percolate after the next step of movement to Spec,CP₂, allowing CP₂ to be attracted to the matrix clause.

We consider next a final option, in which we find successive CP movement: CP₃ moves to the specifier of CP₂, followed by movement of CP₂ to the specifier of CP₁. This would derive the sentence in (49), judged by speakers to be ungrammatical (in contrast with the somewhat degraded (48)). Under a feature percolation account, this configuration would involve successive instances of percolation: first from the wh-word to CP₃, next from CP₃ to CP₂. A QP analysis, on the other hand, is unable to derive this structure. As discussed in detail in Cable (2007), Q is a focus-sensitive operator; more than one Q per question word would result in a semantic clash.

The table in (50) presents a side-by-side comparison of the possible analyses of the four different derivations considered in (46)–(49). Crucially, under the assumption that Q may select for either DP or CP in Ch’ol, we correctly derive the three possible structures in (46)–(48). Given the impossibility of merging multiple Qs in a construction with a single wh-element, we immediately rule out the ungrammatical (49).10 Under a percolation account, on the other hand, there is no principled way to rule in the grammatical structures while ruling out the ungrammatical structures. Specifically, a feature percolation account requires (i) that percolation be optional, and (ii) that features percolate from a specifier to the immediately dominating projection. There is then no principled way to rule out the successive percolation which would derive the ungrammatical (49), without the addition of stipulations. For example, one might stipulate that percolation could take place a maximum of one time. This would require the derivation to somehow “count” or keep a record of whether percolation has already taken place, adding unnecessary complexity to the already optional operation.11

We thus argue that clausal pied-piping in Ch’ol provides additional evidence in favour of a QP approach to pied-piping.

We now return to the question of why clausal pied-piping is ungrammatical with irrealis complements. Recall from section 3.2 above that despite no obvious difference in syntactic size, irrealis CPs are aspectually restricted, appearing only with imperfective aspect. We propose that something about the defective nature of irrealis CPs makes them unselectable by Q. At the moment, this remains a stipulation, though we note it correctly derives the empirical patterns seen so far. We believe a fruitful avenue to pursue would be that there is something incompatible with the semantic type of the irrealis. Arregi (2003: 141), for example, argues that pied-piped CPs must reconstruct at LF, and that this requirement could be derived “from a strong condition on the interpretation of traces, which only allows them to be interpreted as variables over individuals.” If irrealis clauses are not semantically individuals, then we would correctly rule out the possibility of pied-piping. We leave a full semantic analysis for future work. We note that the restriction that Q in Ch’ol selects for wh-DPs or (realis) CPs correctly rules out pied-piping of the non-finite embedded clauses examined in section 2.2 above.

Finally, we argue that a QP approach provides a better means to account for cross-linguistic variation in the availability of clausal pied-piping than a feature percolation account, and further, that Ch’ol provides some important data points regarding the typology of this construction. Specifically, we turn here to the question of why clausal pied-piping is possible in Ch’ol and Basque, but not in English and Spanish. Ortiz de Urbina (1993: 216) suggests two possible correlations, neither of which work for Ch’ol. At the time of Ortiz de Urbina’s writing, the descriptions of clausal pied-piping involved SOV languages, Basque and Quechua, which he suggested as a possible correlation. Verb-initial Ch’ol obviously problematizes this possibility. A second, and more promising possibility, Ortiz de Urbina notes, is that while wh-words may not cooccur with overt complementizers in English and Spanish, they are obligatory in Basque embedded clauses. Specifically, Ortiz de Urbina (1993) suggests that the null complementizer in languages like English and Spanish is subject to constraints governing empty categories, following Stowell (1981), ruling out clausal pied-piping. Again, however, Ch’ol shows us that this story can’t be right at a cross-linguistic level: the complementizer che’ is ungrammatical in both long-distance wh-extraction and clausal pied-piping environments (see e.g., (22) above). The variation between Ch’ol and Basque in this respect suggests this should not relate to a deep property of clausal pied-piping.

Under a QP analysis, variation in the availability of clausal pied-piping is attributed to variation in the selectional properties of the Q head (in addition to restrictions noted in Cable 2007). Thus, while we do not arrive at an answer to why Q selects CP in Ch’ol but not in English, selectional variation is independently needed. Under a feature percolation account, however, variation must be attributed to the availability of a core syntactic operation, a path we suggest—following extended discussion in Cable (2007) and subsequent work—adds unnecessary complication to the grammar.

Finally, though this paper has focused on clausal pied-piping driven by interrogative wh-words, recall that wh-questions, focus, and relativization all involve movement to the left edge of the clause, and are typically taken to be in competition for a single left-edge position. Focussed elements may also trigger clausal pied-piping with inversion, also described by Arregi (2003) for Basque. Specifically, the question in (51a) is felicitously answered by the sentence in (51b), in which the focussed subject has undergone secondary movement to the edge of the embedded clause, followed by pied-piping of the embedded clause to the matrix clause. Leaving the focussed element in situ within the fronted clause in (51b) results in ungrammaticality (unless a pause is present, in which case the configuration can be understood to involve two sentences; see fn. 7).

Relativization may similarly trigger clausal pied-piping, at least for some speakers. An example of long-distance extraction for relativization is shown in (52a). Some speakers found the option with clausal pied-piping in (52b) good, while others found it degraded.

We suggest that this variation could be related to a processing issue, having to do with complexity of the construction, but we save further investigation—as well as a more detailed look at focus and relativization in the domain of pied-piping—for future work.12