3.1 Wh questions in Khalkha Mongolian

Khalkha Mongolian is a wh-in-situ language. In simplex sentences such as those in (8), the wh phrases may remain in their base positions. In each example, a dedicated wh-question particle be/we appears at the clausal periphery and marks the clause as a wh question.

In embedded-clause constructions, the wh-question particle be/we marks its clause as a wh question and indicates wh scope:

In (9a), the Q particle is located inside the embedded clause, and the matrix clause lacks a Q particle. The sentence can only be interpreted as having an embedded wh interrogative; it cannot be interpreted as a matrix-scope question. In contrast, in (9b), the Q particle is located in the matrix clause, and the embedded clause does not have a Q particle. The sentence can only be interpreted as a matrix-scope wh question.

The remainder of section 3 investigates island effects arising from LDS out of embedded interrogatives. A distinction is drawn between two types of movement: ordinary LDS, defined as the clause-external scrambling of a non-wh constituent, and wh LDS, the scrambling of a wh constituent. This distinction is motivated by the empirical fact that the two operations exhibit differing sensitivity to wh islands in Khalkha Mongolian. Data will show that ordinary LDS may proceed from an embedded interrogative, whereas wh LDS from the same domain exhibits weak-island effects.

3.2 Ordinary LDS out of wh islands

It has been observed cross-linguistically that LDS seems to be able to escape wh islands (Zemskaja 1973, Saito 1985, Yoshimura 1992, Abe 1993, Bailyn 1995, Bošković & Takahashi 1998, Tanaka 1999, Kim 2003, Miyagawa 2005, Ko 2018, Ko et al. 2019, and Bailyn 2020). This apparent lack of wh-island effects is attested in Khalkha Mongolian. Consider the following example.

In (10), the embedded object DP (ter nomig ‘that book’) has undergone LDS out of the embedded wh interrogative. Native-speaker consultants report that the sentence is acceptable under the intended reading. To further confirm the pattern, (11) is an example involving a ‘whether’ island.

In (11), the embedded object DP (ter nomig ‘that book’) has undergone LDS out of the embedded ‘whether’ clause. The result in (10) is replicated: consultants report that the sentence is acceptable under the intended reading.

In addition, time and place adverbials, which can undergo LDS quite easily, may scramble out of wh islands. This point is illustrated by the following two examples.⁷

As shown in (12) and (13), LDS of either time or place adverbials is acceptable out of embedded wh questions.

These examples illustrate that although LDS cannot proceed out of strong islands (section 2), it seems to be able to proceed out of wh islands.

3.3 Wh LDS out of wh islands

A wh phrase, too, can be scrambled out of an embedded wh interrogative. Depending on the construction, there is fine gradience in acceptability;⁸ nevertheless, speakers generally accept the following type of construction with ease. Notably, the LDS-ed wh phrase receives an embedded-scope interpretation, exhibiting an obligatory-scope-reconstruction or radical-reconstruction effect in the sense of Saito 1989.

Wh LDS displacing an argument out of a declarative clause is also allowed. Consider the following examples. In (15a), the in-situ wh argument yamar nomig ‘what kind of book’ takes matrix scope. Scrambling this wh phrase out of the embedded [–Q] CP is acceptable, as indicated in (15b).

Next we turn to the crucial data of wh LDS involving non-arguments. While certain adverbials can be scrambled clause externally as demonstrated earlier, if the adverbial that undergoes scrambling is a wh phrase, the sentence becomes ungrammatical. Two examples, (16) and (17), illustrate this fact. In (16), a locative wh phrase (khaana ‘where’) taking embedded scope is scrambled out of the embedded interrogative, landing in the matrix-initial position. Judgments are uniform across all speakers consulted: it is not possible to interpret this dislocated wh phrase as a part of the embedded clause. Similar results are obtained when the wh phrase undergoing scrambling is a temporal wh phrase (khezee ‘when’), shown in (17).

Note in addition that it is not the case that khezee ‘when’ and khaana ‘where’ can never be scrambled. Both wh phrases can be scrambled locally within a clause or clause externally so long as they do not cross a wh island (or other islands). Example (18) illustrates that these two wh adverbials can undergo intermediate scrambling to the sentence-initial position.

These wh adverbials can also be scrambled long distance out of declarative CPs. Examples illustrating instances of LDS of khezee ‘when’ and khaana ‘where’ are the following.

In these examples, the embedded CP is not an interrogative. The wh phrases khezee ‘when’ and khaana ‘where,’ taking matrix scope, may proceed out of these embedded CPs. Although the surface orders of these two examples are mildly degraded (indicated with ?), for all native speakers consulted, they sound far better than (16) and (17) under the intended interpretation.

3.4 Summary of the patterns

The full range of patterns presented so far is summarized in table 1 with representative example numbers.

As the table indicates, ordinary LDS is essentially unconstrained, with ordinary LDS of adverbials being somewhat degraded, something also observed in other languages (Bailyn 2020). Since ordinary LDS of adverbials is nevertheless easily accepted by the native-speaker consultants, for the purpose of the current investigation I will assume that Khalkha Mongolian permits adverbial LDS, at least to a certain degree, and leave the source of such mild degradation to future work.

The data further suggest that wh LDS patterns differently from ordinary LDS. Although LDS of wh arguments is more or less acceptable out of a wh island, wh LDS of adverbials is completely ungrammatical out of a wh island. This pattern is an instance of the argument–non-argument asymmetry, mirroring the familiar weak-island effects observed with A′ movement such as English wh movement. The pattern additionally bears a resemblance to a previous observation made by Huang 1982: although ‘when’ and ‘where’ pattern with ‘who’ and ‘what’ in being quite freely extractable at LF, ‘when’ and ‘where’ pattern more like true adjuncts in terms of narrow-syntax A′ extraction—they are subject to a stricter locality requirement and cannot be easily A′ extracted out of wh islands (Huang 1982: 534–543; see also Tsai 1994).

3.5 The nature of the asymmetry

If wh LDS indeed has similar syntactic properties as regular A′ movement, the fact that it exhibits the asymmetry observed in table 1 is expected. This sub-section investigates in greater detail the nature of that asymmetry, wrapping up section 3.

4.1 Diagnosing LDS landing sites through thematic topics

Khalkha Mongolian LDS cannot freely target an arbitrary landing site in the matrix clause. In all the examples we have examined so far, LDS lands in the matrix-initial position, preceding the matrix subject:

It is known that cross-linguistically, LDS that does not target such a position is often degraded. For example, LDS that targets some position below the matrix subject, as schematized in (32), is noted to be significantly degraded in Mongolian compared to scrambling all the way to the matrix-initial position (Gong 2023b: 24–25). This contrast is also attested in other scrambling languages such as Hindi (Bhatt 2003 and Dayal 2017).

Assuming for now that there is a specific landing site or sites that LDS must target, I adopt the following basic template for the Khalkha Mongolian extended clausal periphery, taking a cartographic approach (based on Rizzi 1997, Rizzi 2013a, and Saito 2013).

In embedded clauses, C hosts the complementizer gej. In interrogative clauses, Force hosts the Q particle, which can either be the wh-question particle be/we (in a matrix question or in an embedded question) or the yes/no-question particle uu/üü (which appears in matrix questions in the data reported in this article). ThemeP hosts thematic topics, headed by the topic marker bol, and is projected “as needed” (in the sense of, e.g., Rizzi 1997). For expository purposes, the templatic structure in (33) further assumes that ThemeP is located below ForceP, following previous work both on potentially universal fine CP structure (Rizzi 1997 and Rizzi 2013a) and on typologically similar languages like Japanese (Saito 2013).

A few additional notes are needed in order to clarify the nature of thematic topics in Mongolian and justify the presence of a dedicated projection for them in the periphery. In this language, topics are often indicated by the marker bol. A bol-marked phrase can be interpreted as a “theme” (i.e., an aboutness topic or thematic topic, as in Kuno 1975) or as a contrastive topic. Therefore, a sentence like (34) is potentially ambiguous between thematic- and contrastive-topic readings of the subject.

Importantly, although a contrastive reading is readily available for bol-marked phrases in most cases, the thematic reading is possible only when a phrase is located in the clause-peripheral position. In (35), bol marks the object ter nomig (‘that book’), which is not in the sentence-initial position. Here, the object can only receive the contrastive reading.

The thematic reading of the object in (35) becomes available if the bol-marked object linearly precedes the subject and occupies the sentence-initial position:

In Gong 2023a, thematic topics in Mongolian are analyzed as follows. Theme, the head that hosts the thematic topic, is located in the extended clausal periphery and requires its specifier to be filled. This explains the fact that thematic topics must be located in the clausal periphery and cannot be in situ in their base-generated positions. As a functional head, Theme does not probe, and a DP may receive thematic-topic interpretation only if it occupies spec-ThemeP (e.g., Miyagawa 2010). Spec-ThemeP is similar to a criterial position in nature (Rizzi 2006, Rizzi 2017, and Rizzi & Shlonsky 2007): a DP receives an interpretation associated with topicality when it occupies this position. Additionally, thematic topics are not limited to root clauses in Mongolian. As noted in Gong 2023a, finite CP complements of the verb ‘to say’ allow embedded thematic topics:

The embedded subject Bat is marked by bol, giving rise to an embedded thematic-topic reading. The co-indexation ensures that the sentence is not a direct quotation (Vermeulen 2013). Only embedded clauses with a large enough structure such as a finite CP can accommodate thematic topics. Complement clauses with smaller structures (e.g., embedded non-finite clauses) disallow thematic topics.

From this brief background we see that (i) ThemeP serves as a fixed “landmark” within the extended clausal periphery and (ii) ThemeP can appear in both root clauses and embedded finite CPs. I will leverage these two properties of ThemeP to diagnose the landing site of two types of LDS.

First, intermediate scrambling can target a position above the bol-marked thematic topic, but LDS cannot. Recall a previous example with a subject thematic topic:

The thematic reading of this subject topic remains accessible even when the subject no longer counts as the sentence-initial constituent after intermediate scrambling applies. In (39), ter nomig (‘that book.acc’) has undergone intermediate scrambling to the sentence-initial position, preceding the bol-marked subject Bat. The thematic reading remains accessible.

The availability of the thematic reading in (39) suggests that intermediate scrambling can land in a position above the thematic-topic position. In contrast, LDS may never precede a thematic topic. Consider the following example, where the matrix subject Yör is marked with bol and receives the thematic-topic interpretation, as indicated in the translation.

Crucially, LDS targeting a position above this thematic-topic matrix subject Yör bol is not possible. The word order in (41) is out.

Is it possible for LDS and matrix thematic topicalization to co-occur at all? The following example, though still quite degraded, presents a contrast with (41) in that speakers report that the word order is plausible.

Given (41, 42), I adopt the working assumptions that LDS (i) targets a fixed position in the matrix clausal periphery and (ii) can in principle target a position below the thematic topic.

4.2 Diagnosing LDS landing sites through relative clauses

Evidence from relative clauses (RCs) further clarifies the nature of the clausal periphery and the specific landing sites of LDS. In particular, the behavior of LDS within RCs suggests that wh LDS targets a higher landing site than ordinary LDS. After establishing the necessary background on Mongolian RC structure and related analytical assumptions, I demonstrate how these positions are structurally differentiated.

5.1 The structure and interpretation of wh in situ in Khalkha Mongolian

As mentioned in section 3, Khalkha Mongolian is a wh-in-situ language: in a wh question, the wh phrase does not move to the clausal periphery to take scope. In simplex sentences, wh phrases remain in their base-generated position. The periphery is occupied by a Q particle, yielding a question interpretation:

Following previous work, I assume that an in-situ wh phrase in Khalkha need not undergo covert movement at LF to its scope position and is instead interpreted in situ (Baker 1970, Nishigauchi 1986, Nishigauchi 1990, Cheng 1991, Aoun & Li 1993, Tsai 1994, Chomsky 1995, Reinhart 1998, Tsai 1999, and Kratzer & Shimoyama 2002, among others). The in-situ wh expression is bound by a phonologically null operator, which undergoes narrow-syntax movement from its base-generated position near the wh expression to the scope position spec-ForceP (Watanabe 1992). The schema in (58) models this null-operator movement following Tsai 1999. Under this approach, the in-situ wh phrase is interpreted as a variable bound by the operator that is moved to the scope position.¹⁶

The existence of movement of a null operator is supported by evidence from the “additional-wh” effect (Watanabe 1992 and Hagstrom 1998; cf. Richards 1998). In Khalkha Mongolian, an in-situ wh phrase is degraded when it appears inside an embedded interrogative:¹⁷

But when a second wh phrase is added outside of the island to make a multiple question, the sentence improves:

Addition of a wh phrase inside the wh island does not lead to improvement:

In Tsai 1999’s analysis of such effects, which builds on Watanabe 1992, exactly one null operator moves from an in-situ wh expression to its scope position in narrow syntax. At LF, the wh question is interpreted via the operator unselectively binding the wh expressions in situ as variables. In-situ wh-island effects like the one in (59) occur when the operator associated with the only wh expression (‘what’ in (59)) is forced to move from its base position inside the wh island to the scope position hosted by the matrix Q particle. Adding another wh phrase outside of the wh island repairs the effect: in (60), the null operator can move from the wh expression external to the wh island to the scope position in the matrix clausal periphery. This option is not possible in (61), since both wh expressions are inside the island.

As mentioned above, the landing site of this operator (i.e., the scope of the wh expression) is closely connected to the position of the Q particle in wh questions. In a simplex sentence, the operator simply moves to the matrix spec-ForceP headed by the Q particle. In a construction where a wh expression in an embedded clause takes matrix scope, such as (62), the operator undergoes successive-cyclic movement from the base-generated position in the embedded clause to the specifier of the matrix ForceP, passing through the intermediate CP edge, as shown in (63).

5.2 A derivational approach to locality in LDS

This section provides the analysis for the locality patterns in the two types of Khalkha LDS. Building on Kitahara 2000, Kitahara 2002, and Saito 2003, I implement a derivational approach to scrambling chains. I assume that a movement chain is created via copy and deletion (Chomsky 1993) and an element undergoing movement in syntax carries with it a collection of features. An overt DP, for example, has a phonetic feature P and a feature D that enables it to have reference and participate in binding/coreference relations (Saito 2003). Furthermore, the properties of the head (e.g., γ) of a landing site (spec-γP) can determine which feature(s) of the moved phrase are retained or deleted at that landing site for interpretation (e.g., operator features are retained at operator positions). Meanwhile, there are also certain features that drive narrow-syntax movement and do not directly interact with LF interpretation (e.g., EPP).

To see how this works, consider English wh movement in a sentence like Who did John see?. The wh expression moves overtly from its in-situ position to spec-CP. In addition to being a DP, the wh expression who is equipped with an operator feature O_wh, which enables the wh expression to be interpreted as an operator once it is moved to the specifier of a CP headed by an interrogative complementizer. In addition, it possesses a separate wh feature. The positing of this wh feature is motivated by the idea that in successive-cyclic movement, the intermediate C heads (and potentially other phase heads), despite not being interrogative complementizers themselves, are equipped with a corresponding feature that drives movement of an embedded wh phrase to their specifier (e.g., Chomsky 2000, McCloskey 2000, and McCloskey 2002). This wh feature is assumed to be semantically inert, that is, deleted once it is satisfied, prior to LF interpretation. Thus, a simple wh-movement derivation would be modeled as follows.

The movement of who carries with it a set of features {P, O_wh, wh, D}. The P feature is retained at the landing site of the chain and deleted at the tail, deriving the fact that this is an instance of overt movement. Assume further, following Saito 2003, that deletion is constrained by the properties of the landing site, such that features remain in positions where they are valued or selected. In (64), the feature O_wh is retained at spec-CP because this is an operator position headed by an interrogative C that can value the O_wh feature on the wh phrase.¹⁸ The movement of who is driven by the wh feature, which deletes prior to LF interpretation. Finally, the D feature is selected at the complement position of the verb see and is neither valued nor selected at spec-CP. Therefore, D is deleted at the head of the chain and retained at the tail.¹⁹

This approach derives the island effects in scrambling examined here. Following among others Rizzi 1990, Chomsky 1995, Rizzi 2001, and Rizzi 2004a, I assume that the wh-island effect in wh LDS and the lack of it in ordinary LDS should be handled in terms of Relativized Minimality. The core intuition behind Relativized Minimality is that a local structural relation between two elements X and Y is blocked if a third element, Z, intervenes between them and Z is a potential bearer of the relevant relation:

While early formulations of Relativized Minimality distinguished broadly between heads and phrases and whether a phrase is in an A or A′ position. (Rizzi 1990), subsequent developments have introduced a more granular hierarchy of features to account for more complex A′ interactions. Specifically, Abels 2012 (see also Starke 2001 and others) argues that syntactic features are organized into sub-class–super-class relationships. Under this approach, a moving element carrying only a general super-class feature is blocked by any intervening element of that same class. An element carrying a more specific sub-class feature may bypass a super-class intervener, meanwhile acting as an intervener for all elements within both its own sub-class and the broader super-class (Abels 2012: 248).

Building on Bailyn 2020, I implement a version of this approach for the current analysis and suggest that the classification needed to derive the LDS facts in Khalkha Mongolian is the following.²⁰

Building on Grewendorf & Sabel 1999, Kawamura 2004, and Bailyn 2020, I assume the Khalkha featural inventory contains a scrambling feature Σ. This feature is responsible for driving ordinary LDS.²¹ While Σ drives syntactic movement, it does not directly interact with LF and deletes prior to interpretation. In the classification (66), Σ is an independent class from what I label as A′. A′ has a sub-class Op, which contains operators (similar to quantificational elements in the formulation of Rizzi 2004a). O_wh is a member of the sub-class Op, which entails that it is also A′. In contrast, the wh feature only belongs to the super-class A′. This organization is compatible with the idea that O_wh, unlike wh, has an operator-like behavior that has consequences for semantic interpretation.²²

The featural organization in (66) makes two predictions. First, movement involving Σ will not interact with any A′ features, since the two are independent of each other. Second, an O_wh intervener will block movement involving wh. Both predictions are borne out by the LDS patterns in Khalkha Mongolian.

5.3 Deriving ordinary LDS and wh LDS

Consider the case of ordinary LDS. It is allowed to escape wh islands and does not exhibit an argument–non-argument asymmetry, because it is driven by Σ and thus may freely proceed across the A′ intervener imposed by the embedded interrogative:

Let us see the derivation for this example step by step:

First, the embedded wh interrogative is built. As shown in (68a), the null operator Op moves from the in-situ position to spec-ForceP, establishing the scope of the wh question. This null Op carries with it the O_wh feature, which enables it be to interpreted at spec-ForceP. The O_wh feature is deleted at the base position. Next, the matrix clause is built. Since ordinary LDS targets matrix spec-AspP (section 4), I suggest that the probe with the Σ feature that drives LDS is hosted on Asp in Khalkha. Once AspP is introduced at the matrix-clause level, ordinary LDS applies to ‘that book,’ a phrase that also has a Σ feature. As shown in (68b), the scrambled DP ‘that book’ carries with it P, D, and Σ features. P is retained at the landing site, deriving the fact that this is overt movement. D is deleted at the landing site and retained at the base position, because D is selected at the object position of the embedded verb ‘bought.’ Σ deletes after driving the LDS of ‘that book’ to its landing site, prior to LF interpretation. Although ordinary LDS crosses Op at spec-ForceP of the embedded interrogative, it does not interact with Op due to their different featural specifications: ordinary LDS involves Σ feature, whereas the Op has O_wh, a type of A′ feature.²³ This configuration is schematized in the following tree.²⁴

Let us turn now to wh LDS. Unlike ordinary LDS, wh LDS is sensitive to wh islands, displaying a selective locality effect sometimes characterized as an argument–non-argument asymmetry. Rather than resolving the ongoing debates regarding the source of this asymmetry (see footnote 12), the current goal is to leverage the phenomenon to probe the syntactic properties of wh LDS. Under the current analysis, the wh-island sensitivity of wh LDS follows from Relativized Minimality: the dependency established by wh LDS is of the same featural type as the operator being crossed at the specifier of ForceP. As an example, consider the ungrammaticality of non-argument extraction in wh LDS:

The configuration is illustrated as follows, assuming that wh LDS is driven by the wh probe hosted on F.

The attempted wh LDS, illustrated with dashed arrows, brings the non-argument wh expression to the matrix clause, making a stop at the edge of the intermediate CP due to CP being a phase. Crucially, wh LDS involves the feature wh, a member of the A′ class. This movement is blocked by the intervening operator at spec-ForceP, carrying the O_wh feature. Since O_wh belongs to Op, it is also a member of the A′ class; therefore, movement of wh crossing an intervening operator is ruled out.

5.4 Further predictions and discussion