The syntax of wh-phrases, narrow foci, and neg-words in Georgian

2.1 Headedness

Georgian is a head-final language, though not consistently so, in contrast with prototypical head-final languages like Korean and Japanese. On the one hand, numerous constructions provide evidence for head-finality. Post-positional phrases (3), genitive+noun combinations (4), participial relative clauses (5), and small clauses (6) are strictly head-final:

Similarly, object+verb idioms are preferably verb-final, and, for some speakers, lose their idiomatic meaning in VO, as shown in (7) (see also Skopeteas and Fanselow 2010).

In contrast, other projections exhibit head-initial properties. E.g., while complementizer placement in Georgian is often flexible, there are no clause-final complementizers, (8), which provides evidence against a head-final CP.

There is also a modal unda ‘have to, must’, which, in contrast to finite verbs in Georgian, can only be found clause-medially and not clause-finally, (9). This suggests that it is found in a higher, head-initial projection, such as AuxP.^[4] The CP- and AuxP-related facts are contrary to what would be expected in a strictly head-final language. I take this to mean that the clausal spine in Georgian above the VP is head-initial, which makes Georgian clausal syntax similar to that of German (Haider 2010), except for the lack of V2.

At the same time, there is considerable flexibility with respect to the order of elements within the VP, with both VO and OV widely attested, as shown in (10). Both word orders are frequent in discourse and can be found in all-new contexts (e.g., Tuite 1998: 42). Most authors agree that OV is underlying (Aronson 1990; Boeder 2005: 64; Harris 2000: 141; McGinnis 1997a, 1997b; Nash 1995; Pochkhua 1962; Skopeteas and Fanselow 2010). This view is adopted here as well.

This, however, raises questions about the syntactic underpinnings of the frequently attested VO. Both OV and VO in Georgian can be information-structurally neutral. Preverbal and postverbal direct objects (DOs) do not differ in their definiteness or specificity. The fact that VO orders are possible as neutral contexts points to their syntactic organization: as argued in Neeleman (2015), neutral word orders are derived by X⁰-movement (which is always leftward), as opposed to phrasal movement to the right. Accordingly, I take Georgian VO to be derived by short (e.g, V⁰-to-v⁰) movement of the verb (as opposed to e.g., displacement of the DO to the right). This is in line with Skopeteas and Fanselow (2010), who also derive neutral Georgian VO by verb raising, and emphasize that this head-movement is semantically vacuous. Accordingly, the derivations of the examples in (10) are schematized in (11):

The availability of this kind of verb movement means that there is a suitable head position on the left side of the clausal spine for the verb to move to, which fits well with the analysis of the clausal spine above the VP as head-initial. Furthermore, deriving VO via verb movement means that the two object positions in OV and VO orders are one and the same syntactic position, which explains lack of interpretational differences between the two.

2.2 The position of the verb in verb-final clauses

There is evidence that the verb in Georgian does not leave the VP in OV clauses (or the vP in VO clauses). The tests commonly used for determining the position of the verb are based on the relative scope of verbal negation and elements such as NPIs and quantifiers associated with verbal arguments (Han et al. 2007; Simpson and Syed 2014, a.o.). The condition that these tests rely on is that the exponent of negation and the verb form a constituent and, were the verb to move, the exponent of negation would move too – e.g., after cliticizing to the verb. This condition is obtained in Georgian. Verbal negation in Georgian is merged and interpreted low in the clause, below the base position of the external argument. The exponent of negation is a clitic on the verb and cannot be separated from it by any other material, including adverbs, as in (12) (similarly to Romance languages; Cinque 1999):

This serves as evidence that the negative exponent and the verb combine into a single unit that cannot be broken up. This can be achieved either if negation is a head and combines with the verb (via syntactic head movement, m-merger, or a combination of both), or if negation is phrasal/adverbial and cliticizes to the verb. To avoid postulating an unusual architecture with a low NegP projection, I take Georgian negation to be adverbial, and therefore phrasal, and a clitic in the phonology; see also Erschler (2015: 47).

Now, let us consider the interaction of a quantified subject and a negated verb, with the quantified subject preceding the verb, as in (13) below. There are two possible readings that such a clause can have, depending on the relative scope of the two elements: numeral > negation or negation > numeral. The availability of the negation > numeral reading would be indicative of the negation + verb complex (covertly) raising past the subject to a higher position, while the availability of numeral > negation alone would indicate lack of such movement. In Georgian, a quantified subject scopes over verbal negation, which suggests that negation is generated and interpreted below the position of the subject and, consequently, that the negation + verb complex does not raise past it from its low position in the VP/vP. In particular, (13) can be used to describe a group of three students in which one or two students won’t know the answer (numeral > negation). On the other hand, it cannot be used to say that it is not the case that fewer than three (out of three) students know the answer (*negation > numeral). Note that, because (13) is a VO-clause, the verb has undergone V⁰-to-v⁰ movement, but the scope facts demonstrate lack of further movement.

Similarly, the position of the negation + verb complex can be diagnosed by using a verbal argument that contains disjunction. A disjoint reading is only predicted to be felicitous if the disjunction scopes over negation (disjunction > negation), whereas a conjoint reading should be available if the disjunction scopes below negation (negation > disjunction) (Shibata 2015). This is shown in (14):

In Georgian, in contrast with English, when a direct object contains a disjunction, it scopes over verbal negation, which is manifested by the availability of the disjoint reading, and the unavailability of the conjoint reading, as shown in (15). The means that the verb is generated and interpreted below the position of the disjunction.

The evidence from these two tests indicates that the verb does not raise from its base position in broad-focus declarative clauses. I assume that verbal morphology is assembled via m-merger as opposed to head movement (Harizanov 2014; Matushansky 2006).^[6] Accordingly, I conclude that verbs in Georgian stay low in the clausal structure.

2.3 Case marking

Case marking of verbal arguments in Georgian varies between nominative, ergative, and dative, depending on the tense category of the verb (known as ‘series’ in the Kartvelological tradition). This is illustrated in Table 1. Because case-related facts in Georgian are complex, and in-situ case licensing plays an important role in the argument, these facts are discussed in detail below.

With respect to the structural positions and case licensing of verbal arguments, I follow the existing proposals by Legate (2008) and Nash (2017). According to both, case licensing in Georgian is achieved in situ – though, depending on the series of the verb, arguments are generated in different projections. The two analyses differ in some important respects – e.g., in treating ergative case as dependent (Nash 2017) or inherent (Legate 2008). Furthermore, Nash’s (2017) account is a hybrid one, in that some cases are assigned by functional heads and others via dependent case assignment (Baker and Vinokurova 2010; Bobaljik 2008; Marantz 1991). I refrain from proposing a dedicated approach to case licensing in Georgian and, instead, summarize Legate (2008) and Nash (2017) below: most importantly for the present purposes, they align on the structural positions of nominals, while diverging in the mechanics of case assignment. Nothing in the current proposal depends on the latter.

In the present series, subjects (transitive and intransitive) carry nominative, as shown in (16). According to both Legate (2008) and Nash (2017), nominative case is licensed by T⁰ via c-command. Evidence for T⁰ as the licenser comes from the regular absence of nominative in the absence of T⁰: e.g., masdar nominalizations, which do not have a TP, cannot have a nominative argument, as shown in (17). The nominative argument, therefore, is always found in the scope of T⁰, though its thematic role and structural position vary by series. A dedicated Voice⁰ head (Kratzer 1994; cf. Nash’s 2017 Event⁰ head) is also merged above the vP in the present series. It combines two functions: thematically licensing the external argument of vP and expressing the eventuality with respect to the reference time in T⁰. The nominative subject is merged in Spec,VoiceP – and case-marked by the c-commanding head T⁰. In transitive clauses in the present series, the direct object is located in the VP and is assigned dative case: either by v⁰ (Legate 2008) or Voice⁰ (Nash 2017).

The structural positions of the arguments in the present series are illustrated in (18), based on (16b). An intransitive subject would likewise be found in Spec,VoiceP in the present series.

Next, in the aorist series, ‘active’ subjects carry ergative, and ‘inactive’ subjects and objects carry nominative. In the absence of Voice⁰ (Event⁰) in ergative constructions, T⁰ directly takes the vP as its complement (other functional projections, like AspP, notwithstanding). If the verbal domain contains one argument, its case is checked against T⁰ and is nominative, as illustrated in (19). This is the case for unaccusative verbs in the aorist series.

For Georgian unergatives, I am adopting Nash’s (2017) covert causative analysis, in the spirit of Hale and Keyser (1993), which groups them together with transitives. Both are illustrated in (20). In transitive contexts in the aorist series, (20b), where two arguments need to have their case checked, the lower one of the two, the direct object, receives nominative from T⁰.

Ergative, according to Nash (2017), is assigned in Spec,vP via the dependent case mechanism. It is assigned in the absence of VoiceP (EventP), which would have allowed the subject to receive nominative from T⁰. Instead, ergative appears in configurations where T⁰ ‘sees’ two arguments with unvalued case features in the same vP domain and marks the higher one with dependent case. The presence of the ergative is therefore a sign that VoiceP is absent and the dependent case algorithm tracks the higher argument. Legate (2008) also takes ergative to be assigned in Spec,vP but treats ergative as inherent case. Accordingly, an illustration of a transitive clause in the aorist series (with an ergative subject) is provided in (21), based on the example in (20b):

Finally, in the perfect series, ‘active’ subjects carry dative, and ‘inactive’ subjects and objects carry nominative. The evidence that dative subjects are true subjects comes from binding: dative subjects can serve as antecedents for tav-reflexivization, which in other series is a property unambiguously characteristic of subjects (Harris 1981: 117; McGinnis 1997a; Thivierge 2021; van der Vijver et al. 1995). Thivierge (2021) takes perfect series verbs to be (dyadic) unaccusatives; accordingly, they lack VoiceP (Kratzer 1994). Following Thivierge’s (2021) analysis of Georgian dative subjects, based on the distributional properties of verbal agreement and the formation of causatives, I take dative subjects to be generated in Spec,ApplP (the same position as dative indirect objects in transitive clauses; Lomashvili 2011). This is shown in (22) and (23) below. Being case-licensed by Appl⁰, dative subjects do not enter the dependent case calculation.

To recap, subjects and objects in Georgian vary in their case marking and structural positions based on the series of the verb, but they do not leave the vP/VoiceP domain of the clause for the reasons of case assignment – instead, they are case-licensed in situ. This means that Georgian does not have movement for case assignment.

5.1 Non-focal material: distribution

Topicalized constituents in Georgian appear in the left and right peripheries of a clause. In terms of their interpretational properties, there are two types of topics: contrastive/aboutness topics, which are found in the left-periphery, and familiarity topics, which may be found either in the left periphery or postverbally, as backgrounded/given material. Contrastive/aboutness topics either introduce or change the main topic of the utterance.^[9] Familiarity topics, in turn, refer to given/backgrounded discourse material but cannot introduce new referents (Frascarelli and Hinterhölzl 2007; Givón 1983; Şener 2010). In the left periphery, contrastive/aboutness topics typically precede familiarity topics, and both types of topics necessarily precede wh-phrases/narrow foci in the IPrP.

The topical status of a constituent can be diagnosed in the following way. The appearance of a new (contrastive) topic may result from deliberately replacing an element in a contrast set: the explicit juxtaposition of two constituents attests to the presence of a contrast between the two. The availability of a contrastive reading can only obtain with topicalized or focused constituents (Lambrecht 1994); therefore, if the same clause also contains a narrowly focused constituent, the constituent with a contrastive interpretation must be a topic. To illustrate, in (37a), the contrastive topic Giorgi is substituted by another one, Mariamma (cf. Neeleman and Van de Koot 2008; Şener 2010). Note also that the contrastive topic Mariamma obligatorily precedes the narrowly focused constituent xaʧ’ap’uri ‘khachapuri’, as manifested by the ungrammaticality of (37b):

I do not assume dedicated positions for topical constituents, in line with the work that shows that topics do not occupy syntactic positions available only for a particular topic type (Neeleman et al. 2009; Zwart 2007). Instead, I adopt the view that left-peripheral topics are housed in the CP projection, while the preference for contrastive/aboutness topics to precede familiarity topics is an IS-property that is not directly rooted in syntax. In the absence of dedicated projections, this ordering results from the communicative preference to present the main topic or contrastive material first, followed by backgrounded/familiar material (cf. Zwart 2007; Neeleman and Van de Koot 2008; Neeleman et al. 2009). I take the housing projection to be CP as opposed to e.g. TP, given that there is no explicit evidence in Georgian that topics may be available in CP-less structures (cf. Iatridou and Kroch 1992).

There is conflicting evidence with respect to whether left-peripheral topicalized constituents come to occupy their positions by movement or base-generation. This issue requires further scrutiny, which goes beyond the scope of this paper. Nothing in the current account hinges on the mechanisms that underlie displacement of topical/given material.

5.2 Neg-words: distributional properties

Neg-words (also referred to as n-words; Laka 1990) in Georgian (in broad-focus contexts) are also required to appear in the IPrP (Aronson 1990: 47). Placing neg-words further to the left of the verb results in ungrammaticality (regardless of the thematic role of the neg-word). This is shown in (38) and (39):^[10]

Direct object neg-words, such as e.g., araperi ‘nothing’, for many (though not all) speakers, can occur postverbally as well, as shown in (40a). Neg-words other than direct objects, in contrast, are considerably more resistant to postverbal placement, as shown in (40b–c).

The obligatory IPrP-placement of neg-words (except when co-occurring with wh-phrases & foci – more on this in Section 5.4) makes their distribution similar to that of narrow foci and wh-phrases. This is not surprising, given that there is a robust semantic connection between these types of constituents: focusing and questioning pick a particular entity from a set of alternatives, while neg-words “eliminate entire sets of contextual alternatives” (Drubig 2003: 15). The requirement for immediately preverbal placement of neg-words, found in Georgian, also has some cross-linguistics parallels. In Iron Ossetic, an Iranian language of the Caucasus that has been influenced by Kartvelian, neg-words are also obligatorily immediately preverbal; in contrast with Georgian, no neg-words can be placed postverbally in Iron Ossetic (Borise and Erschler 2023; Erschler 2010, 2012, 2013). In Hittite, an extinct Anatolian language, neg-words together with relative pronouns, wh-phrases and indefinites formed a cluster that had to be adjacent to the verb, either by preceding or following it (Huggard 2015; Sideltsev 2014, 2016, 2017).

5.3 Neg-words: non-topical status

One of the main syntactic properties of neg-words in Georgian is their inability to displace into the clausal peripheries. The reason for this is the non-referential status of neg-words: since they refer to empty sets and do not pick out any referent, they cannot act as topics.^[11] This is shown in (41) and the exchange in (42) (the felicitous non-IPrP placement of the neg-words in [42] is discussed in detail in Sections 6.5 and 7).

In avoiding topical interpretation, neg-words in Georgian pattern together with their counterparts in other languages, such as Italian, in which they similarly avoid topicalization in Clitic Left Dislocation (CLLD) structures (Alexiadou 2006), as shown in (43):

5.4 Neg-words: in-situ placement

Based on the examples in Section 5.2 and 5.3, the following generalization emerges: given their non-referential nature, neg-words cannot receive a topical interpretation and displace into the left periphery. At the same time, because the appearance of post-verbal backgrounded material is also a type of topicalization (familiarity topics), right-peripheral placement of neg-words is also excluded. Based on these generalizations, and the fact that non-IS-motivated movement, such as movement for case, does not exist in Georgian, the conclusion is that Georgian neg-words are always found in situ. As such, they can provide useful evidence about the structural positions of wh-phrases and narrow foci, as shown in Sections 6.5 and 7.

The in-situ status of neg-words is also supported by some general syntactic tests, such as island effects and interaction with adverbs. Because these tests are also used in Sections 6 and 7 to assess the placement of wh-phrases and narrow foci, their results, when applied to neg-words, are provided here for the sake of the uniformity of the argument. First, neg-words can be embedded in islands, as shown in (44), which indicates lack of movement to the left periphery (Spec,CP) of the embedded clause:

Second, it is possible to show that, e.g., object neg-words surface low, below AspP-level adverbs like often/seldom (Cinque 1999), which also supports their in-situ status:^[12]

Before turning to using neg-words as a diagnostic for the position of other constituents in a clause, recall from Section 5.2 that some speakers allow for direct object neg-words in the postverbal domain:

At first sight, it might seem that this violates the generalization derived above, that neg-words are necessarily found in situ. This is only an apparent problem, however. In Section 2.1, it was shown that VO is possible as a neutral word order in broad-focus contexts, and that it is derived by short movement of the verb, which means that the direct object in VO orders, like in OV ones, is found in situ. Accordingly, it is not surprising that direct object neg-words may be found in the postverbal domain: such placement corresponds to their in-situ position, after the verb undergoes movement to v⁰. Postverbal object neg-words, then, behave like any other postverbal objects. Also, because neg-word direct objects are allowed in the postverbal domain, an alternative analysis of obligatory adjacency between neg-words and verbs – one based on the presence of a NegP projection, the specifier of which would host the neg-word, with the verb raised to Neg⁰ – is moot. Such an analysis would not explain the dual behavior of direct object neg-words.^[13]

6.1 Preview of the results

To recap from Section 3, wh-phrases, if compared to narrow foci (and neg-words), have the simplest distribution: they can only occur in IPrP, as illustrated by (47); This is why wh-words are often picked to exemplify the behavior of constituents with the IPrP-requirement, since their distribution is the most consistent one.

By extension, preverbal occurrences of narrow foci (and neg-words) may be assumed to have the same underlying syntax, while also, for some independent reason, allowing for more flexibility. However, the remainder of this paper shows that this is not the case, since wh-expressions in Georgian differ in their syntactic properties from both narrow foci and neg-words: they undergo A-bar movement to the specifier of a dedicated projection between the vP/VoiceP and TP, accompanied by raising of the verb to the head of the same projection (WP). The schematic representation of the structure underlying WHQs is provided in (48), based on (47a):

The movement analysis for wh-phrases is a type of a Spec-Head configuration that creates adjacency between the preverbal element and the verb. The alternative would be for the wh-phrase to stay in situ, and for any material intervening between the wh-expression and the verb to be displaced. A range of evidence supports the Spec-Head configuration as underlying WHQ-formation in Georgian: island effects, weak crossover (WCO) effects, and interaction of wh-expressions with adverbs and neg-words; these tests are discussed in detail in the next subsections, where the behavior of wh-phrases is compared to that of preverbal foci.

More specifically, island and WCO facts show that wh-question formation involves movement of the wh-expression, though it does not tell us whether the movement is overt or covert (since island effects and WCO effects are sensitive to both overt and covert movement). Next, the interaction of wh-expressions with adverbs and neg-words shows that the movement that wh-phrases undergo is overt but does not identify the height of its landing site. Finally, the interaction of wh-expressions and interrogative complementizers shows that the landing site for wh-constituents is above the thematic domain but below the CP-area.

Similarly to wh-phrases, narrow foci, if found in the preverbal part of the clause, obligatorily surface in the IPrP. This is shown in (49) for narrow focus in a reply to a WHQ, in (50) for contrastive focus, and in (51) for a constituent modified by only.

As with wh-phrases, there are two possible ways that the adjacency between the focused constituent and the verb in (49)–(51) may be derived in: the movement-based Spec-Head configuration, and the in-situ adjacency, made possible by the displacement of intervening material to the right or left periphery. The tests applied in the next subsections to both wh-phrases and preverbal narrow foci provide evidence for the in-situ placement of narrow foci in the IPrP. The syntactic structure that they motivate is shown in (52), based on (49):

In the following Sections 6.2–6.5, the behavior of preverbal foci and wh-phrases is compared with respect to island effects, WCO effects, and distributional effects relative to adverbs and neg-words; in Section 6.6, the remaining arguments for the positioning of wh-phrases are provided.

6.2 Island effects

Island effects are robustly present in wh-questions in Georgian, which can be demonstrated with a relative clause (RC) island, shown in (53a), alongside a grammatical version that does not involve a wh-phrase in (53b).

Evidence of this kind is commonly taken to mean that a wh-phrase raises to a position in the left periphery of the clause that it is merged in – e.g., because it is attracted by a [+Q] feature on the interrogative C⁰ (Chomsky 1995 and subsequent work). This movement is overt in English but covert in wh-in-situ languages like Mandarin Chinese (Huang 1982). When a wh-phrase is embedded in a RC, this movement is blocked – presumably, because the Spec,CP position of the embedded clause is already occupied by the relative pronoun. Island effects, however, are found both with overt and covert wh-movement to the left periphery and do not differentiate them – they only signal the attempt at either kind of movement to the periphery of the clause in which the wh-phrase is merged. Therefore, (53a) does not yet tell us about the location of overt wh-phrases in embedded clauses – it would be compatible both with an in-situ wh-phrase that raises to CP covertly or one moving to the CP overtly.

Next, let us consider preverbal narrow foci when embedded in a RC island: (54) and (55) illustrate corrective focus and only-focus in RCs, respectively.

The grammaticality of these examples, in contrast with (53a), provides evidence against movement of narrow foci to the left periphery: if such movement were involved, placing foci inside a strong island, like a RC, would result in ungrammaticality.

6.3 Weak crossover effects

The next diagnostic to consider is weak crossover (WCO) effects. WCO configurations can help distinguish in-situ constituents from those that have undergone A-bar movement – e.g., in a Spec-Head configuration – or LF movement. WCO effects in languages like English, where wh-expressions undergo overt A-bar movement to Spec,CP, are thought to result from the wh-expression crossing a variable that it is coindexed with on its way to Spec,CP (Chomsky 1976: 19; Higginbotham 1980; Koopman and Sportiche 1982; Reinhart 1983; Safir 1984). In (56a), a WCO effect between a coreferential pronominal within the subject and a wh-object is illustrated; the same effect does not obtain in the absence of coreference between the pronominal and a wh-object, as in (56b). In wh-in-situ languages, there is no overt ‘crossing’, since the wh-expression does not leave its base position, but WCO effects may still be present – arguably, due to the LF movement of the wh-expression over the variable (cf. Huang 1982; Aoun and Li 1993 for Mandarin Chinese). In Georgian, the equivalent of (56a), with the intended coreference, is ungrammatical, as shown in (57a);^[14] in the absence of coreference, the example is grammatical, as in (57b), in parallel with (56b).^[15]

This indicates that the wh-phrase crosses a coindexed pronominal on the way to its landing site. Furthermore, the word order in (57a), parallel to the English example, suggests that wh-phrases in Georgian undergo overt movement.

Preverbal narrow foci do not give rise to WCO effects in the same configuration, which provides evidence against a movement analysis, in contrast with wh-phrases. This is shown in a (58a) for corrective focus, and in (58b) for only-focus:^[16]

6.4 Interaction with adverbs

Relative placement of a constituent with respect to an adverb whose height is independently known – e.g., based on Cinque’s (1999) hierarchy of adverbs – is frequently used to determine constituent height. Wh-phrases and preverbal narrow foci in Georgian behave differently with respect to AspP-level adverbs like often/seldom (chosen here due to their mid-way height on the clausal spine): wh-phrases are most naturally placed higher than/to the left of iʃviatad ‘seldom’, and preverbal foci lower than/to the right of it:

These data also testify to the movement of wh-phrases and verbs, and in-situ placement of preverbal narrow foci.

6.5 Interaction with neg-words

Finally, we should look at the interaction of wh-phrases, narrow foci, and neg-words. As shown in Section 5.2, like wh-phrases and narrow foci, neg-words have a requirement to surface in the IPrP. At the same time, we have seen that neg-words necessarily stay in situ, since they cannot receive a topical interpretation and undergo displacement into a clause-peripheral position. Because of this, neg-words in wh-questions and narrow focus contexts provide evidence about wh- and focus placement. As shown below, the relative distributions of wh-phrases and narrow foci with respect to neg-words are different.

For wh-phrases, the interaction with neg-words provides further support for the (overt) movement analysis. First, all neg-words, regardless of their argument/adjunct status, obligatorily surface postverbally in WHQs. (60) shows that neg-word direct objects must surface postverbally in WHQs with wh-subjects:^[17]

In a parallel fashion, a neg-word subject must be placed postverbally in a WHQ with a wh-object, as shown in (61).

The fact that the wh+verb complex can only surface to the left of a (postverbal) subject neg-word, as shown in (61), combined with the fact that neg-words are found in situ in Georgian, suggests that the wh-expression and the verb surface in derived positions, as schematized in (62). If so, wh-phrases must occupy these derived positions in a Spec-Head configuration.

Moving on to preverbal narrow foci, consider first a context with a narrowly focused subject and a neg-word direct object. As shown in (63a), the neg-word araperi ‘nothing’ cannot precede a narrowly focused subject, because that would involve a derived position of the neg-word, which is disallowed (in contrast, a referential NP in this context can precede the narrowly focused subject). Leaving araperi ‘nothing’ in its base position, on the other hand, is possible, as shown in (63b). Note that the neg-word is postverbal in (63b) because neg-objects behave like other direct objects, as discussed in Section 5.4: they can be preceded by a verb that has undergone V⁰-to-v⁰ movement.

The neg-word staying in situ here does not yet provide definitive evidence for the structural position of the focus and the verb: both an in-situ placement of focus (accompanied by V⁰-to-v⁰ movement of the verb, which derives postverbal placement of the neg-word), and movement of both elements, the focus and the verb, to WP would result in the same linearization, (63b). With that in mind, a context with a narrowly focused DO and a neg-word subject, as in (64), is more informative. Here, the subject neg-word aravin ‘no-one’ can precede a narrowly focused DO. In contrast, postverbal placement of aravin in the same context is degraded – even though such placement would mirror the word order in the WHQ, as was shown in (61a):^[18]

The picture that emerges from (63) and (64) contrasts with what we have seen for wh-questions in (60) and (61), where neg-words, regardless of their thematic role, cannot linearly precede wh-phrases. The contrast in (64) is particularly telling: given that the neg-word is in situ, narrow focus and the verb also must be structurally low. Since it would be unexpected to find a discourse (focus) projection below the base position of the subject, I take the narrowly focused object and the verb in (64a) to be in their in-situ positions. Note that the felicitous responses in (63) and (64) correspond to the two neutral, broad-focus word orders in Georgian, with the arguments staying in situ – SVO and SOV, respectively. This is expected on the analysis that narrow foci, just like neg-words, do not move: if these two elements co-occur, the resulting word order can only be identical to the neutral one. The derivations of (63b) and (64a) are provided in (65a) and (65b), respectively.

These facts contrast with the relative distribution of wh-phrases and neg-words: regardless of their thematic role, wh-phrases always ‘win over’ the IPrP, with neg-words surfacing postverbally, as was shown in (60) and (61). These facts would be incompatible with an in-situ analysis of wh-phrases and can only be accounted for if wh-phrases undergo movement.

The idea that narrow foci in Georgian stay in situ is further supported by the finding that some speakers allow for direct object neg-words to intervene between a focused subject and the verb, as shown in (66).^[19] The resulting word order, then, corresponds to the unmarked SOV order, and means that both the focused subject and the neg-word object are found in their base positions:

In contrast, the same word order (narrow focus – neg-word – verb), is impossible when the theta roles are reversed – that is, with a subject neg-word intervening between a focused direct object and the verb, as shown in (67). This is expected under the current proposal, since in this word order neither verbal argument is found in situ:

The interaction between narrow foci and neg-words, therefore, provides decisive evidence against a uniform structural treatment of preverbal narrow foci and wh-phrases in Georgian. It shows that preverbal narrow foci in Georgian are interpreted in situ. Wh-phrases, in contrast, undergo A-bar movement to Spec,WP, accompanied by head-movement of the verb to W⁰.

6.6 Height of the target projection for wh-phrases

The last remaining step is to determine the height of the WP projection, to which the wh+verb complex raises. To do that, note that in WHQs with a neg-word subject, like (61a), the wh+verb complex is located higher than the subject position. The subject position, depending on the series of the verb, is either Spec,vP (for ergative subjects), Spec,VoiceP (for nominative subjects), or Spec,ApplP (for dative subjects), as discussed in Section 2.3. Accordingly, in WHQs with a neg-word subject, the movements that the wh-object and the verb undergo take them to a projection above the base position of the subject.

Next, evidence that wh-movement targets a position below the CP in Georgian comes from embedded wh-questions. In embedded wh-questions, the optional interrogative complementizer tu, a C⁰, precedes the wh-phrase. As (68) shows, tu cannot follow a wh-phrase, which means that the wh-phrase is located below the CP (see also Erschler 2015: 62).

Let us bring together what we know about the height of wh-phrases’ landing site. The interaction with adverbs provides evidence for overt A-bar movement of wh-phrases, which I take to be driven by a syntactic interrogative feature [+Q]. Island effects and WCO effects signal the presence of further (covert) movement of the wh-phrase/its subpart to the CP-domain, presumably required for its correct interpretation. Interaction with neg-words shows that wh-phrases move above the position of the subject, but interaction with interrogative complementizers shows that they land in a projection below CP. I propose that this projection is WP, located on the top of the VoiceP, as was shown in (48). Accordingly, wh-phrases do not raise all the way to the left periphery in Georgian. Lack of wh-movement to the left periphery explains why there is no cross-clausal wh-movement, as was shown in (28): wh-phrases simply do not raise high enough (overtly) in their clause to move into the higher clause.