On wh-movement in echo-questions and crosslinguistic variation

This paper examines echo wh-questions, a rather understudied phenomenon even in extensively described languages such as English. In particular, it focuses on a particular type of echo questions, such as those made in response to a previous declarative (e.g., –Mary said {mumble}./ –Mary said what?) or a previous wh-question (e.g., –Who said {mumble}?/ –Who said what?). Such structures are examined from a comparative perspective, analyzing data from three different languages regarding Multiple wh-Fronting: English vs. Russian, with attention to Spanish. On the one hand, this paper considers the key, cross-linguistically common features of echo questions and discusses their underlying derivational structure. On the other hand, contrary to the standard assumptions that echo questions necessarily require wh-in-situ, this paper focuses on the availability of different options of overt echo wh-movement among the languages under consideration. It is argued that in echo questions, similarly to what happens in canonical interrogatives, wh-movement proceeds successive-cyclically and is subject to parametric variation. 66 Beyond Philology 18/3


Introduction
This paper aims to shed some light on the syntactic behavior of echo wh-questions (henceforth wh-EQs), which are interrogative sentences produced as an immediate response to a previous utterance requesting information about some portion of the stimulus that has been missed or unheard. This type of question is exemplified below, in (1b) (hereafter, the echo-introduced wh-phrases appear in small caps): (1) a. Mary had tea with Dracula.
b. Mary had tea with WHO? [from Sobin 2010:132] As stated in Sobin (2010:132), "EQs present considerable challenges to theories of interrogative syntax predicated on the behavior of non-EQ interrogatives", as they behave in a quite unusual way. For instance, EQs appear to counterexemplify some general statements about the formation of true, non-EQ questions such as the obligatoriness of wh-movement or the sensitivity to Superiority effects, which are illustrated in (2) for English: (2) a. Whati did John say ti? b. Who1 said what2? c.* What2 did who1 say?
However, English wh-EQs are immune to the obligatory whmovement and the consequent verb raising, as we have seen in (1b) (cf. (2a)). Moreover, when an EQ has more than one whword, Superiority effects can be easily violated without resulting in ungrammaticality (see (3b); cf. (2c)): (3) a. What did Dracula drink at Mary's party? b. What did WHO drink at Mary's party? [from Sobin 2010: 132] For these and other reasons, EQs tend to be seen as a nonsyntactic phenomenon, as they systematically disobey the general rules of question formation. It seems "unprofitable to attempt to integrate them into the analysis of the more usual types of questions" (Culicover 1976:73), because "the grammatical rules of the language should not generate them" (Cooper 1983:149). However, here I agree with Sobin (2010: 131), who made the first attempt to capture English EQs in terms of generative syntax, arguing that EQs, being "in the realm of 'automatic' and 'untutored' knowledge", are "of great interest and relevance to analyses of question formation". Different from true, canonical questions, EQs do not request for new information: "instead they are requests for confirmation of something someone has heard" (Carnie 2006:340) or has not understood. EQs are strongly bound to the previous discourse and, thus, sometimes are referred to as backward citations (Escandell 1999) or reprise questions (Ginsburg and Sag 2000). To illustrate this point, observe again the previous example, (3). The questioner in (3b) cannot hear a part of the wh-question in (3a) pronounced by their interlocutor (suppose, they cannot hear Dracula). So, the speaker formulates a wh-EQ, in which the echo wh-word who substitutes the unheard portion of the utterance in (3a), the rest of the utterance being reproduced without changes, including the wh-word what.
So, echo wh-words are referential items, in the sense that they ask about a referent which has been already mentioned in the immediately previous discourse. By using as a question an undeformed utterance (i.e., a question with wh-in-situ), the speaker presents themself "as being unable to complete the utterance in a satisfactory way" (Fiengo 2007:76) and asks the addressee to repeat a missing bit of language and to assign a value to the echo wh-word. 1 1 Of course, EQs can be produced in immediate response to an utterance not only in order to request for repetition, as in (2b), but also to express speaker's surprise, as in (i) (following Bartels (1997), I call the former type unheard EQs and the latter, amazement EQs): (i) a. A: We're going to Pakistan on vacation.
(English) b. B: You're going WHERE on vacation?! c.
A: Well, the nature is beautiful there.
[from Šimík 2009: 5] In this paper, I restrict my attention only to unheard EQs, which can be considered interrogative constructions both from syntactic and semantic points of view (i.e., they seek to reduce the speaker's ignorance about some missed portion of the stimulus, denoted by a wh-word, under which the In this paper, I mainly restrict my attention to the wh-EQs containing two wh-items, like those in (3b), in which one whword is "inherited" from the utterance and the second one is echo-introduced. Moreover, I put forth some novel evidence supporting a view that EQs are syntactic phenomena, underlined by a particular syntactic structure: namely, one involving two CP-levels (see Sobin 2010;Chernova 2015Chernova , 2017. As it will be argued, such a view allows accounting for several striking properties of EQs without appealing to purely discursive notions. Here I focus on some key properties of EQs across two languages with different wh-fronting strategies: English and Russian. Additionally, I bring into discussion Spanish wh-EQs, in order to support some points of my argumentation. Consider the examples below. As is well-known, Russian, (4c), differs from English and Spanish, (4a) and (4b) respectively, in that the former exhibits obligatory Multiple wh-Fronting (hereafter, MWF) in questions with more than one wh-word. In other words, all wh-items must undergo movement in Russian (as generally in Slavic; see Rudin 1988;Richards 2001;Bošković 2002; among many others): (4) a. Who1 sees whom2? b. Spanish ¿Quién1 ve a quién2? who.NOM sees who.ACC 'Who sees whom?' c. Russian Kto1 kogo2 vidit? who.NOM who.ACC sees 'Who sees whom?' proposition contained within the utterance is true). In contrast, the meaning of an amazement EQ is rather similar to an exclamative: in (ib), the speaker B knows exactly what has been said, however, in their opinion, Pakistan is the least expected place to go on vacation.
Recall that the few existent studies on EQs have argued that this type of interrogatives does not exhibit overt wh-movement (see Noh 1998;Iwata 2003;den Dikken 2003;Fiengo 2007;Sobin 2010;among others). Consider the English example in (5), where the questioner cannot hear a part of the previous utterance, a wh-question, (5a), and asks for clarification. Observe that in the EQ the echo wh-word who can only appear in-situ (notice that the apparent Superiority violation does not lead to ungrammaticality in this example), (5b) vs. (5c). 2  As shown in (6), in Russian, in addition to the wh-in-situ option, (6b), the echo wh-phrase (here, kto 'who.NOM') has two other possible landing sites. On the one hand, it can undergo partial wh-movement (Fanselow 2005) 3 to some preverbal position, below the wh-word "inherited" from the utterance (here, kogo 'who.ACC'), (6c). On the other hand, the echo wh-word can also appear at the leftmost position, above the utterance's wh-word, (6d). 4 Finally, apart from English and Russian EQs, I also consider here some data from Spanish, a language with restricted availability for multiple wh-movement under certain licensing contexts (see Etxepare and Uribe-Etxebaria 2005;Uriagereka 2005;Gallego 2017). Observe from the examples below that in this language there are at least two available options for the echo-inserted wh-word: 5 3 Following Fanselow's (2005:439) terminology, I assume that "movement is partial whenever the phrase has been displaced but its final landing site is below the relevant position". 4 As reported in Chernova (2015), for some Russian speakers the in-situ position of the echo wh-word, as in (5b), is dispreferred over any other option with movement. There is also some variation regarding the leftmost position of the echo-introduced wh-phrase, (5d): many speakers judge it as marginal although possible, while others consider it perfectly acceptable; finally, few speakers reject it. 5 Notice that the EQs in (6)  Similar to what we have seen for English, Spanish speakers show a strong preference for the wh-in-situ option, (7b), and unanimously judge as ungrammatical the possibility of overt echo wh-fronting into the leftmost position, (7d). However, differently from English and similarly to Russian, in Spanish, the echo wh-item can undergo partial movement into some immediately preverbal position, below the wh-word "inherited" from the previous utterance, (7c). 6 Let us summarize the data seen so far. On the one hand, all three languages under consideration allow the in-situ option in wh-EQs (although, it is dispreferred in Russian). On the other hand, partial wh-movement to some immediately preverbal position is allowed both in Spanish and Russian, but it is blocked in English. Finally, overt echo wh-movement to the left edge of the interrogative clause is acceptable only in Russian. This is summarized below in Figure 1.

Figure 1
Echo wh-movement across languages In this paper, I mainly discuss why languages like Russian and Spanish do allow wh-movement in EQs and other languages like English do not. This paper aims to propose a syntactic account that can capture the attested parametric differences regarding movement. 7 As already mentioned, I focus here on EQs with two wh-words: one comes from the stimulus, and the other one is echo-introduced. I propose that similar to any type of syntactic movement, echo wh-movement proceeds successive-cyclically. However, its legitimacy is restricted by certain well-established parametric differences among wh-fronting languages. Namely, I argue that it depends on two main factors: (i) the clause-typing properties of the echoed utterance (declarative, interrogative, etc.) and (ii) the number of escape hatches out of phases (such as CP) available in a particular language. 8 I show that we can get a deeper understanding of even such a striking (and apparently 'non-syntactic') phenomenon such as EQs if we analyze it comparatively (here, between three typologically different wh-fronting languages) and under a unifying theory: namely, Cable's (2010) Q-based approach to the interrogative syntax. The paper is organized as follows. In section 2, I review several key properties of wh-EQs, which distinguish them from canonical wh-questions, and propose that EQs have a particular syntactic structure. In section 3, I discuss three theoretical assumptions I make to account for the echo-puzzle: namely, Sobin's (2010) proposal on the double-CP structure underlying this type of questions; then Cable's (2010) Q-based approach to the derivation of canonical wh-questions; and, finally, the idea that languages can differ concerning what portion of the structure becomes a phase domain (see Gallego 2007Gallego , 2010den Dikken 2007, among others). Afterward, in section 4, I offer an account that allows us to capture the attested variation regarding echo wh-movement in a uniform manner. First, I address EQs with full echo wh-extraction, to the leftmost position of the clause (available only in Russian, (6)), and then I consider EQs resorting to partial wh-movement (allowed in Russian and Spanish, (6)-(7)). Section 5 concludes this paper.

Key syntactic properties of wh-EQs
Many of the challenging properties of EQs have been previously reported in the literature (e.g., see Sobin 1978Sobin , 1990Sobin , 2010Parker and Pickeral 1985;Dumitrescu 1992;Noh 1998;Ginsburg and Sag 2000;Escandell 1999Escandell , 2002Iwata 2003;Fiengo 2007;Sudo 2007;Vlachos 2012;Chernova 2013Chernova , 2015Chernova , 2017among others). Here I consider only a few of them, the most relevant ones for the topic at hand. 9 As already mentioned in the previous section, perhaps, one of the most well-known and, at the same time, striking features of EQs is that they preserve the clause-typing properties of the sentence they "echo" (see Sobin 2010;Noh 1998;Escandell 2002; among others). As we have already seen for English in (5) (consider also (8) below), when an EQ repeats a previous wh-question, it has to maintain the wh-interro-gative morphosyntactic features of the echoed utterance (here, the fronted what and the raised auxiliary did). In addition, the EQ introduces its own syntactic features (here, the echo wh-word who). Although the resulting structure clearly violates Superiority, the structures in (5b) and (8b) are grammatical contrary to their counterparts in (c). Similarly, the EQ in (9b), based on a previous polar question, has to preserve the yes/no nature of the echoed utterance, (9a); so, the echo wh-word must remain in-situ: However, observe that in the case of "echoed" declarative sentences, as in (10), the echo inserted wh-word can either appear in-situ, as in (10b), or undergo overt wh-movement into the left periphery of the question (with a consequent raising of the auxiliary did), as shown in (10c) Generally, an EQ can reproduce any kind of utterance: for example, an exclamative, (11), or an imperative, (12): (11) a. U: What a great pleasure this is! b. EQ: What a great WHAT this is? (12) a. U: Go to see the archaeologist.
b. EQ: Go to see WHAT/WHO? As the reader may observe, the strategy of "echoing" is broadly always the same: an EQ repeats the stimulus and replaces the unheard portion by a wh-word. 11 Interestingly, the interrogative clause-typing of the EQ itself co-occurs with the clausetyping of the echoed sentence. In Escandell (2002), this echoproperty is called mood clashes, while in Sobin (2010) it is named Comp freezing. In both cases, the terminology seeks to capture the fact that the resulting EQ conserves the clausetyping markers of the echoed utterance. Interestingly, an echo-introduced wh-word has always the widest possible scope, independently of its position inside the clause (see Sobin 2010;Chernova 2015Chernova , 2017. That is, as shown below for English, independently of whether the echointroduced wh-phrase appears in the root clause, (8)-(12), or deeply embedded, (13b), it always receives wide scope and seeks for an answer: Crucially, the root-scope phenomenon is also observed in EQs with two wh-items, as in (8b). As is well-known, canonical, non-echo multiple questions presuppose exhaustification of each quantifier, giving rise either to pair-list or single-pair readings (see Higginbotham and May 1981;Hagstrom 1998;Krif-ka 2001 Likewise, in an EQ based on a previous wh-question (see example (15) below), only the echo wh-word (who in (15b)) receives scope, while the wh-word inherited from the previous utterance (what) requires no response. 12 In fact, the only appropriate answer to (15b) is (15c) (as far as the agent of the action described by the stimulus in (15a)  Moreover, observe that an echo wh-word can appear both inside strong and weak islands (without consequences for the grammatical status of the correspondent sentence). Again, just as in previous examples, it necessarily receives wide scope. This is illustrated below for English, where the island effects are created by sentential subjects, (16), adjuncts, (17) Finally, observe that there is an interesting piece of evidence suggesting that EQs do "actively involve syntax" (Sobin 2010:135): the echoed utterance and the correspondent EQ may show different person-agreement features and deictic elements. That is, the content of an EQ is sensitive to the changing discourse roles between the speaker and the addressee (see Dalrymple and Kaplan 2000;Harley and Ritter 2002). This is illustrated below for Spanish: The deictic accommodation (reflected on the verb and the pronouns) corresponds to the two dependents of the participants of the speech act: speaker and addressee (1 st and 2 nd person, respectively). These changes are unexpected if we assume that EQs are simply a type of direct quote (e.g., Mary said: "I am hungry"). Rather, EQs seem reminiscent of indirect questions (e.g., Mary said that she was hungry).

EQs as a double-CP structure
In this paper, I argue that, despite the appearance of being a purely pragmatic phenomenon, unaccountable under any syntactic rule, EQs actually do "actively involve syntax" (Sobin 2010:135). Some of the previously discussed echo-properties, such as co-existence of syntactic features of two different clause-types and wide scope for the echo-inserted wh-word independently of its position within a clause, suggest that wh-EQs are structurally different from true wh-questions. 14 Here I assume a particular echo-structure, originally proposed in Sobin (2010): EQs possess their own, interrogative C head (CEQ), in addition to the C head involved in the derivation of the echoed utterance (CU). As a result, the syntactic structure of EQs involves two different adjacent CP projections: namely, CPEQ asymmetrically c-commands CPU. This is schematically represented below: Under the structure in (20), it is expected that the clausetyping features of the echoed utterance would be preserved in the EQ, through projection of the CPU level. However, it does not mean that the derivation of wh-EQs somehow implies a "frozen copy" of the utterance's CP (contra Sobin 2010). Rather, I suggest that a C head of the same type as the one of the stimulus is merged during the derivation of wh-EQs; consequently, the same type of CP (but, importantly, not the same instance of that CP) is built in the course of the standard, bottom-up derivation. Afterward, an additional functional head is merged into the structure: CEQ. It selects CPU as a sister and projects a higher, discourse-bound interrogative projection. CEQ assigns scope to the anaphoric, echo-introduced wh-word within its c-command domain. 15 The higher projection, CPEQ, is also responsible for the request-for-repetition meaning of the resulting question. As the derivation proceeds, we obtain a double-CP structure, as in (20). Following Chernova (2013Chernova ( , 2017, here I assume that Spanish EQs, especially those reproducing a previous yes/no ques-tion, offer an interesting piece of evidence for the double-CP structure proposal. Consider the example below: Notice that, in addition to the echo-inserted wh-word quién 'who', the EQ in (21b) exhibits two items that are absent from the original stimulus, the yes/no question, in (21a): que 'that' (a quotative marker) and si 'whether' (an interrogative particle). The latter appears only in Spanish EQs based on a previous yes/no question: while the lack of si would be ungrammatical in (21b), its presence is blocked in EQs built on a previous declarative, (22), or a wh-question, (23): QUIÉN? that if what bought who 'What did who buy?' As I argued in Chernova (2013Chernova ( , 2017, si is a phonetically realized instance of the interrogative operator Q, which is merged within the CP level of a yes/no question and is responsible for its interrogative interpretation (see Baker 1970). Q tends to be null in Spanish (and also in English) root polar questions, but it becomes phonetically realized in embedded contexts: as si in Spanish (see Rigau 1984;Suñer 1991;Hernanz 2012); if/whether in English (Baker 1970) or se in Italian (see Rizzi 2001). Thus, in Spanish EQs, as in (21b) (assuming the structure in (20)), si is a phonetically realized Q, merged at the specifier of CPU; in the root context, the same position is occupied by its phonetically null counterpart. 16 This is schematized below in (24). So, the absence of si from the EQs in (22) and (23)  Let us now briefly consider the introductory particle que 'that', which can optionally appear in Spanish EQs, independently of the clause-type of the echoed utterance. 17 I take que as a quotative marker (see Escandell 1999). In EQs, it signals that the speaker partially reproduces ("quotes") the words pronounced by her interlocutor in the previous speech turn. The data suggest that this marker is merged within the CPEQ level: observe that que must always precede si: 16 An anonymous reviewer wonders why in English polar EQs if/whether (at Spec,QPU under my account) is not present, (i), even though it is in indirect, embedded non-echo polar questions, (ii): (i) a. U: Did Mary have tea with { mumble }? (English) b. EQ: (*If/Whether) did Mary have tea with WHO? (ii) John asked if/whether Mary had tea with Dracula. Perhaps the answer to the aforementioned contrast is that the auxiliar verb undergoes v-to-T-to-C movement in English root questions (contrary to their embedded counterparts; see Torrego 2001, 2004) and acts as a sort of Doubly-Filled Comp Filter (Chomsky and Lasnik 1977). On the other hand, it has been argued that in Spanish root questions, the verb does not rise so high (see Gallego 2007Gallego , 2010 and references therein; see also the footnote 22). I leave this issue for future research.
17 In principle, the particle que 'that' is optional in Spanish EQs. However, most of my informants note that EQs sound more natural (and are interpreted more easily as echo) when que is present. This opens an interesting question on which factors affect the degree of optionality of que. However, I leave this issue aside for the present.
if has arrived who 'Has arrived WHO?'

A Q-particle approach to wh-EQs
As for the trigger of wh-movement into the left periphery, here I assume the main insights of Cable's (2010) Q-based theory for true, non-echo wh-questions and extend them to wh-EQs. According to Cable, movement of a wh-word in questions is a secondary effect of Q-movement: in other words, the fronted wh-word is not a scope-bearing operator. That is, the syntactic and semantic relations with the interrogative C are established through the help of a Q-particle, which is merged with a whword (or a larger, wh-containing phrase, XP) in its argument position. In languages resorting to wh-ex-situ, when Q is merged with a wh-phrase, the former takes its sister as a complement and projects its own QP layer, which minimally dominates both items. As a result, the first node endowed with the Q-feature being visible for C is QP. This entails that the attraction of the Q-feature into CP triggers movement of the whole QP (no feature percolation being necessary). This is schematically represented below: (26) Derivation of standard wh-questions with overt wh-movement (adopted from Cable 2010: 38) Extending Cable's original theory to wh-EQs, I argue that the derivation of this type of interrogatives involves three crucial elements: (i) an anaphoric echo wh-phrase, corresponding to the unheard portion of the stimulus, merged at the argument position, (ii) a phonetically null discourse-bound interrogative Q-particle (QEQ), merged anywhere in the tree where it c-commands the echo wh-word, and (iii) a discourse-related interrogative head CEQ. 18 Independently of the merging place of the QEQ, by the end of the derivation, it must move to the scope position of the question, a syntactic universal. I propose that in wh-EQs the scope position for the interrogative operator is in the specifier of CPEQ. Thus, echo wh-movement is triggered by a formal imperfection on the QEQ-particle itself and its need to check its feature [QEQ] with the head CEQ.
In principle, under Q-theory, a Q-particle can be merged anywhere in the tree where it c-commands the wh-word. Following Cable (2010), I propose that in wh-fronting languages the size of a wh-containing constituent XP is restricted by the locality-sensitive Agree operation between the Q-morpheme and the echo wh-word. Regarding EQs, it means that an echo Q-particle must agree with the echo wh-word it c-commands within some local domain.
I suggest that all echo-inserted wh-words (recall their anaphoric/referential nature, as opposed to wh-words of the ordinary question) enter the derivation bearing a valued instance of the [wh]-feature, see (27a). Notice, however, that this [wh]feature on echo wh-elements is different from the standard [wh] on wh-words involved in non-echo questions, as only the former are anaphoric items (hence, they bear a [+anaphoric] feature, which I represent here, for simplicity's sake, with an index EQ). As for echo Q-particles, assuming Cable (2010), I propose that they must bear a bundle of features, see (27b). In addition to the interrogative feature [QEQ], I suggest that such Q-morpheme also carries an unvalued instance of [whEQ], which forces the Q-particle to agree with the anaphoric wh-phrase it c-commands: Such Q/wh-agreement is subject to locality conditions. In particular, as discussed in Cable (2010), agreement cannot cross into islands and separate spell-out domains. Under such contexts, the unvalued feature on QEQ cannot be checked and the derivation fails: (28) Within Cable's Q-theory, wh-ex-situ and wh-in-situ structures result from different merging options of Q and its whcontaining sister XP. These options are summarized below. On the one hand, the Q-particle can take XP as a complement and project a QP; then the whole complex QP undergoes movement into CP, resulting in wh-ex-situ. On the other hand, Q can ad-19 Under this assumption, EQs with wh-in-situ (an option allowed in all three languages under consideration) are derived through the help of a particular instance of QEQ that does not need to undergo agreement with an echo wh-phrase. I do not discuss this option in detail in this paper due to space restrictions. For a detailed account, the interested reader is referred to Chernova (2015). join to XP; such Q does not project and, as a consequence, it undergoes movement into the scope position of the correspondent question by itself, leaving its sister in-situ. These two options are represented in (29a) and (29b), respectively: (29) a. QEQ-projection (wh-ex-situ) b. QEQ-adjunction (wh-in-situ) According to Cable, wh-fronting languages always resort to Q-projection, while wh-in-situ languages like Japanese or Chinese resort to Q-adjunction. However, I propose that even in wh-fronting languages, under particular, discourse-bound contexts such as wh-EQs, a discourse-bound Q-particle can resort to both merging options. In other words, in wh-fronting languages, not all QEQ-morphemes need to project. Certain instances of QEQ can resort to adjunction, although they still require agreement with the echo wh-word within their c-command domain. If the echo Q-particle merges locally, no effect arises at the outcome: after agreement, the echo whword remains at its base position and QEQ undergoes fronting into its scope position on its own. However, if such QEQ is initially merged at long distance from its wh-containing goal XP, it forces the latter to undergo partial wh-fronting from its argument position into the edge of a phase, to become visible for the probe QEQ. This type of QEQ-adjunction, parameterized for Spanish and Russian (which I will discuss later in this paper), is illustrated below: (30) QEQ-adjunction (after Q/wh-agreement)

Parametrized points of spell-out
As it is well-established (since Chomsky 2000Chomsky , 2001, the complement domains of the phase heads, standardly v and C, become opaque for further operations as a result of being transferred to the external systems (the so-called Phase Impenetrability Condition; PIC). 20 In addition, assuming the idea that v-movement results in the extension of "checking domains" (see Chomsky 1986Chomsky , 1995, several studies on phases have argued that points of Spell-out are subject to parameterization (see Svenonius 2000;den Dikken 2007;Gallego 2007Gallego , 2010Pesetsky 2007). In other words, languages can differ as to what portion of the structure becomes a phase domain.
The extension of vP's phasehood in a particular language is parasitic on head movement of v into a higher functional projection since v brings together with it its phasal properties. 20 According to Chomsky's (2000Chomsky's ( , 2001 PIC, once the derivation is done at a given stage, correspondent chunks of structure are spelled-out, thereby becoming inaccessible for the further computation. PIC helps to reduce the computational burden, being a constraint that forces the system to "forget" about transferred portions of the structure. According to Chomsky's (2001:14) version of PIC, the transfer of the complement domain of a phase is delayed until the next phase head is projected; afterwards any further syntactic manipulation of the spelled-out chunk of structure is prohibited: Let us first consider Spanish, a language that resorts to the extension of the phase vP into TP (see Gallego 2007Gallego , 2010, due to "one of the most obvious differences between Romance and English […]: v-to-T movement" (Gallego 2006:47). The author captures the very well-known descriptive distinction between the so-called "morphologically rich" languages (e.g., Romance) and "morphologically poor" ones (e.g., English) in terms of Phase Sliding. Namely, in Spanish (but not in English), TP is a phase. 21 The contrast between these two types of languages is schematized below in (32), where α (within the clear shadowed zone) stands for the edge of a phase; meanwhile, β (the dark shadowed zone) represents the phase domain, which gets transferred to the Interfaces and becomes invisible to the higher syntactic nodes: Gallego addresses many properties of Spanish (and other Romance languages) which, according to his proposal, are the result of TP being a phase: e.g., pro-drop, the fact that subjects can appear both pre-and post-verbally (the formers bearing a topic-like flavor), and the lack of obligatory subject-verb inversion in questions (see (33) Observe from (33), that, contrary to English questions (with obligatory subject-verb inversion), in their Spanish counterparts the subject (here, Celia) can appear both above and below the verb. As discussed in Gallego (2007:129), while the question in (33b) has a standard, out-of-the-blue meaning ('there is a reason x, such that Celia did not call her sister because of x'), the question in (33a), with a preverbal subject, receives a marked interpretation. Namely, it can mean either 'why was it Celia (and not another person) who called her sister?' or else 'why was it (true) that Celia called her sister?'. According to the author, in (33a) the preverbal subject appears at the edge of TP and, as a consequence, acquires a topic-like flavor. This is a plausible outcome under Chomsky's (2001) claim that discourse-oriented semantics is related to phase edges. That is, in Spanish the edge of TP can exhibit certain peripheral properties generally attributed to the "standard" phase heads C and v. 22 As for Russian, another language with rich morphology, it has been argued that it has an additional phasal projection, AspP (see Dyakonova 2009;Chernova 2015). This idea recasts the well-known fact about the richness of aspectual morphology in Slavic, as opposed to languages like Spanish or English. As is well-known, in Slavic languages, aspectual differences are encoded in verbal morphology, particularly, in a large number of aspectual prefixes (see Svenonius 2004). For instance, Russian has a fairly simple system of tense and a quite complex system of aspect, which means that interpretation of the former is mostly determined by the latter (see Borik 2006;Borik and Reinhart 2004). It has been proposed that in Slavic languages, similarly to Romance, the verb undergoes movement, but "it remains relatively low", as it "cannot move as high as T" (Svenonius 2004b:6). Namely, the phase head v moves to Asp (see Svenonius 2004a,b;Ramchand 2004;Bošković 2014). I argue that, as a consequence, the phasal properties of vP extend to AspP, as represented below (compare with (32)): 23 (34) 22 Gallego's proposal captures the sense of Uriagereka's (1995) FP, a projection "sandwiched" between CP and IP and encoding discourse-oriented effects. 23 For a detailed discussion of arguments suggesting v-to-Asp movement and its application to canonical multiple wh-questions the reader is referred to Chernova (2015).
A natural effect of the phase extension is the consequent extension of the phase complement domain, a point that is going to be crucial for our purposes. Recall from (31) that, under PIC, only the head H and the edge of a phase HP are visible for further operations, while the complement of H becomes opaque by being transferred. However, if the phase HP extends to ZP, what used to be the edge of HP turns into the domain of the newly extended phase ZP. As a result, all syntactic objects with unvalued feature(s) are forced to escape the domain of ZP, otherwise, the derivation would crash.
Applying this logic to Russian (and Spanish), I argue that after movement of v into a higher head, phase extension takes place. That is, the edge of vP turns into the domain of AspP (in Russian) or TP (in Spanish). Subsequently, all potential goals or elements with any formal imperfection must be removed from the edge of vP to the edge of the higher, newly constructed phase, in order to be visible for further syntactic operations. In contrast, in English, the phase vP does not extend and the verb remains low (in v). This scenario is schematically represented below for all three languages: (35) In what follows, I argue that the availability of echo wh-movement is parameterized across languages by PIC, as such movement proceeds successive-cyclically out of phase domains through available escape hatches along its path. Partial raising to the preverbal area of echo wh-words, an option available in Russian and Spanish EQs, is a consequence of the extension of the vP phasehood and the need of the echo Q-particle to undergo agreement with the echo wh-word it c-commands. In EQs, further movement into the highest level, CPEQ, proceeds through the edge of the lower CPU, which is a phase in all three languages under consideration. As is well-known, the number of available escape hatches out of phases is also parameterized.

Accounting for the parametric variation on echo wh-movement
As already mentioned through the paper, I claim that EQs in principle allow for both wh-in-situ and wh-ex-situ strategies, with an intermediate option: partial wh-movement (allowed in Russian and, marginally, in Spanish). However, the availability of overt wh-movement into the leftmost position of the question is constrained by the clause-type of the echoed utterance. Importantly, there is a crucial observation in Sobin (2010), which I take as a departure point for my argumentation. Consider again (10), repeated below as (36) with some additional items: Sobin observes that English EQs allow for overt wh-movement, as in (36c), only when the echoed utterance is declarative, (36a). In effect, recall from our previous discussion that echo wh-movement is blocked in other contexts (when the echoed sentence is either a polar, (37), or a wh-question, (38)). In such cases, the only available option for the wh-word is to appear in-situ: Sobin argues that EQs must preserve the syntactic character of the stimulus (under his proposal, by "freezing" the CP of the echoed utterance). Thus, (37c) is ungrammatical because overt wh-movement is not compatible with the yes/no syntax of the stimulus in (37a). Similarly, movement of the echo wh-word who in (38c,d) would break the "frozen" CP layer of the echoed wh-question in (38a). However, as we have seen already in (6) (repeated below with additional items as (39)), this prediction does not hold for Russian wh-EQs, where the echo wh-word can undergo overt movement into the leftmost position even when the CP of the echoed utterance has interrogative syntax. This is exemplified below for EQs built on a previous wh-question: As shown in (39), in Russian the echo wh-word kto 'who.nom' can appear in-situ, (39b) (just as in English), but it can also undergo movement to an immediately preverbal position, (39c), or even to the left periphery of the question, (39d), above the wh-word "repeated" from the stimulus.
We have also seen that in Spanish EQs the option of echo wh-movement is neither completely blocked, although it is more restricted than in Russian (see (40)  In Spanish EQs, in addition to the standard wh-in-situ option, (40b) and (41b), the echo wh-word can also undergo partial wh-movement into some preverbal position, the option (c) in (40) and (41) (similarly to Russian and contrary to English). 24 However, movement into the leftmost position, above qué 'what' in (40d) and si 'whether' in (41d) (presumably into CPEQ), is blocked (as opposed to Russian).
Thus, our data suggest that the standard assumption on the mandatory wh-in-situ for EQs does not hold cross-linguistically. In what follows I offer an account for the overt echo whmovement (available only in Russian) and partial echo whmovement (available in Russian and Spanish) that captures the attested parametric variation uniformly.

Echo wh-movement into the leftmost position
Extending Cable's (2010) Q-based theory to EQs (see section 3.2), I argue that in EQs with the echo wh-word at the leftmost position (e.g., Whoi did Mary have tea with ti?, (36c)), what undergoes movement into CPEQ is not the echo wh-word alone, but rather a complex QPEQ projection, which includes the echo Q-morpheme and the wh-word.
I claim that echo wh-movement proceeds successive-cyclically, through available escape hatches on its way up to CPEQ. Following Chomsky's (2001 et seq.) Phase theory, I assume that internal Merge of the fronted wh-phrase to the highest CP 24 Recall our observation that EQs with partial fronting of the echo wh-word, below the wh-item "inherited" from the echoed stimulus, sound quite weird for most consulted Spanish speakers (see the footnote 6). Moreover, an anonymous reviewer brings to my attention an interesting contrast between the following Spanish examples: (i) a. ?? ¿Que dónde QUIÉN estaba? b. ?(?) ¿Que dónde QUIÉN estaba durmiendo? (ii) ¿Qué dónde estaba (durmiendo) QUIÉN? Although both examples in (i) sound rather odd in comparison with (ii), with wh-in-situ, it seems that the question in (ib), with a "heavier" VP, is slightly better than the one in (ia). Although a detailed account of Spanish data falls aside from the scope of this paper, the contrast with their Russian counterpart is noteworthy. Namely, questions with partial movement of the echo wh-word have a higher degree of acceptability among Russian speakers than among Spanish speakers. I leave this interesting issue for future research. node does not proceed in a unique long leap, but rather occurs through the intermediate landing sites, or escape hatches (i.e., every specifier along the movement path). Assuming the double-CP structure of wh-EQs (schematically represented below in (42)), it is expected that the complete echo wh-extraction has to proceed through the edge of CPU on its way into CPEQ. Under this view, the grammaticality of EQs with full wh-extraction crucially depends on the availability of the specifier of CPU as an escape hatch out of the phase domain. I argue that this is precisely the reason why the clause-type of the echoed utterance (declarative vs. interrogative) plays such an important role for overt echo wh-movement.
When an EQ is based on a declarative utterance, the edge of the phase CPU is left unfilled; so it can act as an escape hatch for an echo wh-word on its way to the edge of the higher CPEQ. I claim that this is the reason why the English example in (36c), with overt extraction of the echo wh-word (repeated below as (43b))  However, as we have seen, complete echo wh-extraction out of interrogative contexts is much more restricted and it is subject to parametric variation: such EQs result completely ungrammatical in English (see (37c) and (38c,d)) and Spanish (see (41d)), but they are licit (although slightly deviant) in Russian (see (39d)). This puzzling crosslinguistic variation follows straightforwardly from the current proposal. It is commonly assumed in the literature that, unlike languages of the English type, Slavic languages, which exhibit obligatory multiple wh-fronting in standard, non-echo wh-questions, resort to multiple specifiers of CP (see Rudin 1988;Richards 2001;Bošković 2002; among many others).
In EQs based on a previous wh-question, the specifier of the CPU is occupied by the wh-word "inherited" from the utterance, as in (38)-(39), or, in the case of Spanish EQs based on a polar question, (41), this position hosts si 'whether'. However, in Russian EQs there is an additional escape hatch at CPU. Thus, it is not surprising that a complete echo wh-extraction into CPEQ, through the edge of CPU, is allowed only in this language. The contrast is schematically represented below, for the English EQ in (38c) and the Russian one in (39d) In this respect, Spanish is similar to English in that it resorts to single wh-movement in true questions with more than one wh-word (see below). Hence, this language does not resort to multiple specifiers of CP: ha visto a quién2? who.NOM has seen who.ACC 'Who has seen whom?' 26 As for the marginal status of Russian EQs with complete whmovement, (38d), I argue that it can be accounted for in terms of Relativized Minimality (since Rizzi 1990), namely its reformulation in terms of sensitivity to the feature-specification of the involved elements (Starke 2001; see also Rizzi 2013): (i) In the configuration [...X ...Z ...Y], a local relation cannot hold between X and Y if Z intervenes and Z fully matches the specification of X in the relevant morphosyntactic features [adopted from Rizzi 2013:179].
Roughly, the echo wh-word in (43a) can pass over the non-echo wh-item because the former is more richly specified (by being [+anaphoric]) than the latter. For a detailed discussion of this issue, see Chernova (2015).
b.* ¿Quién1 a quién2 ha visto? 27 who.NOM who.ACC has seen Thus, the ungrammaticality of Spanish EQs with echo whmovement into the leftmost position, as in (41d), is also expected under the same logic as in (44). Recall our discussion that in these languages, in addition to the phase domains projected by C and v, there is another intermediate functional projection that can act as a phase and, consequently, can exhibit A-bar properties (and host elements undergoing A-bar movement). In Russian, such phasal properties are assumed by AspP and in Spanish, by TP.

Partial echo wh-movement
I suggest that in EQs with partial wh-extraction, there is an echo Q-particle that is merged at distance from the echo whword and resorts to adjunction. This means that such QEQ 27 It has been argued in the literature that under particular pragmatic contexts Spanish can allow multiple wh-fronting (see Etxepare and Uribe-Etxebarria 2005;Uriagereka 2005;Gallego 2017), although, presumably, the lower wh-word does not move as high as the first one: (i) ? No sé quién1 a quién2 ha enviado una carta. NEG know who.NOM to who.ACC has sent a letter 'I don't know who sent the latter to whom' (Uriagereka 2005: 2) does not need to project its QPEQ, thus it does undergo movement into CPEQ by itself to check its QEQ-feature. Assuming Cable's (2010) Q-based theory, such Q can adjoin low (e.g., within vP) or high (e.g., at CPU). I assume that such Q bears an unvalued instance of [whEQ]; thus, it has to undergo agreement with the echo wh-word, bearer of the valued instance of the matching feature. The latter must be visible to the former to be able to agree. Recall that, in principle, the Q-particle can be merged anywhere in the tree. Suppose that QEQ is adjoined low (say, at the edge of vP), as represented in (48a). From this position, it can agree with the echo wh-word, valuing its instance of [whEQ], and then it undergoes successive-cyclic movement into its scope position, the edge of CPEQ. By being adjoined, such QEQ does not pied-pipe the echo wh-word, leaving it in-situ. Suppose, however, that the QEQ is merged high (say, at CPU), as shown in (48b) In (48b), the wh/Q agreement cannot take place, as the goal (the echo wh-word) is within the domain of the extended phase and, hence, it is invisible to the higher probe QEQ. Given that the formal imperfection on QEQ cannot be deleted, such derivation fails.
The data in (46)-(47) suggest that in Spanish and Russian EQs with partial wh-movement the QEQ is merged high, as we see that the echo wh-word (the goal) raises to a preverbal position. I assume that in these languages, in order to escape the extended phase domain and remain visible to the probe, the echo wh-word moves to the edge of TP or AspP, respectively. Notice that the "inherited", non-echo wh-word (at Spec,CPU) cannot intervene between the probe and the goal, as it is specified with a different set of features (i.e., it does not bear [whEQ]). This is illustrated below for the Spanish wh-EQ in (46), and the Russian one in (47)

Move
Once agreement takes place and the QEQ deletes its formal imperfection, it undergoes local movement into the edge of CPEQ, reaching its scope position. Evidently, the option of partial wh-movement is not available in English, as this language does not resort to the extension of the vP phase. Thus, the echo wh-phrase remains low, in-situ.
The successive-cyclic nature of echo wh-movement is schematically represented below (the shadowed zones represent the additional host positions and escape hatches for the echo wh-word that are available in Russian and Spanish, but absent from English): 28 A reviewer wonders what happens if the Q-particle is merged at the very end of the derivation, when CEQ is in the structure. In fact, under Cable's (2010) Q-theory, the Q-particle can be merged anywhere in the tree from where it c-commands the wh-phrase. As we have seen, if the Q is merged vP-internally and projects a QP, it triggers explicit wh-movement into the left periphery. However, the Q-particle can also be adjoined directly to CU or CEQ. In such a case, the Q-particle will not project any QP, binding the wh-item at distance; consequently, the wh-word will remain in-situ. In principle, I assume that such derivation is possible for EQs with wh-in-situ (e.g., Mary had tea with WHO?; Did Mary have tea with WHO?; What did WHO drink at Mary's party?; etc.; for a detailed discussion, see Chernova 2015). In this paper, however, I focus on EQs with explicit movement, especially those contexts in which a QP has to circumvent barriers to reach the CPEQ. To sum up, assuming the particular syntactic structure of wh-EQs, the intuition is that echo wh-movement, just as standard wh-movement, proceeds successive-cyclically through the available escape hatches, and it is subject to certain parametric variation. The QEQ-morpheme has to reach its scope position: the edge of the highest phase, CPEQ. Depending on the merging options of QEQ, the echo-introduced wh-word can either be pied-piped into the left periphery of the question or undergo partial movement to the edge of lower phases to be visible for the probe QEQ.

Conclusions
In this paper, I addressed the syntax of wh-EQs in three typologically different wh-fronting languages (English, Spanish and Russian). I argued that echo wh-movement is parallel to standard wh-movement in true questions and is subject to similar kinds of restrictions. Crucially, echo wh-movement also proceeds successive-cyclically, through the available escape hatches on its path. I offered new empirical data showing that in MWF languages (e.g., Russian) the echo wh-item can be fronted into the leftmost position of an EQ independently of the clause-type of the echoed utterance. Meanwhile, in languages resorting to single wh-fronting in multiple questions (e.g., English and Spanish) the possibility of echo whmovement to the left edge of the question is restricted by the type of the utterance: whether it is declarative or interrogative. In addition, I argued that in Russian and Spanish an echo wh-word may also undergo partial movement to a lower, preverbal, position; this option being also parametrically restricted.
Following Sobin (2010), I assumed that wh-EQs have a particular syntactic structure, with two CP levels. As for the nature of the interrogative movement, I assumed the insights of Cable's (2010) Q-based theory and extended it to the derivation of EQs. That is, there is a particular echo Q-morpheme, which is merged in EQs and which regulates the semantics of the echo wh-words; during the derivation, the morpheme has to reach its scope position, the specifier of CPEQ.
On the one hand, I argued that the QEQ-particle may project a QPEQ, which also dominates its sister, the echo wh-word, and pied-pipes it into the left periphery of the question. However, to reach the specifier of CPEQ, such QPEQ has to pass through the lower phase edge, CPU. Thus, it is expected that the syntactic character of the echoed utterance (declarative vs. interrogative) would restrict the availability of the specifier of CPU as an escape hatch. That is, we expect that such movement is allowed in the case of EQs built on a previous declarative utterance, while it is restricted if the echoed utterance is interrogative. In addition, I argued that the typology of wh-movement in true multiple questions (i.e., whether a particular language can make use of multiple specifiers of CP) also determines the final derivational outcome, as it may enable additional escape hatches for extraction of the echo QP.
On the other hand, the QEQ-morpheme may also resort to adjunction and be merged at distance from the echo wh-word. In such case, the latter cannot be pied-piped into the leftmost position of the clause together with the QEQ; however, the whitem still may undergo raising to an edge of a lower phase to remain visible to its probe. Again, this option is also parametrically restricted. I argued that it relies on the mechanism of extension of the vP phase in Spanish and Russian, which has been proposed in the literature on independent grounds.
The two possibilities regarding echo wh-movement attested among the languages under consideration naturally follow from the view offered in this paper. The discussion, hopefully, sheds some more light on the nature of such understudied phenomena as EQs.