2.1 Empirical picture

Personal and locative pronouns in Wolof, exemplified in table 1 and table 2, have a strong paradigm and a clitic paradigm. Locative clitic forms carry a proximal–distal distinction (-i = proximal marker, -a = distal marker).³ There is only one series of object pronouns, for both goal and theme objects (i.e., there is no distinction in Wolof between dative and accusative clitics).

In languages with multiple series of pronouns, as noted by Cardinaletti & Starke 1999, strong and weak/clitic pronouns surface in different environments. This is also the case in Wolof, where only strong pronouns can be coordinated, focused, dislocated, and be complements of prepositions. The object and locative clitic pronouns behave like special clitics in the sense of Zwicky 1977 and almost always occupy a syntactic position different from that of a corresponding non-clitic element. Comparing the strong and clitic pronouns shows that the clitic pronouns are not necessarily prosodically weak/reduced relative to the strong pronouns: the second person plural subject clitic and the second and third person plural object clitics have long vowels and codas. Clitic movement therefore cannot be attributed to the prosodically weak nature of clitics.

I will show that clitics adjoin to the highest head in the extended projection of the verb; I refer to this movement as clitic movement. The behavior of the subject clitic, however, depends on the syntactic position that it finds itself in. If it precedes the highest head in the movement domain of clitics (here argued to be the extended projection of the verb), it surfaces in the initial position and does not necessarily cluster with other clitics (see (2b)). In that case, it still forms a prosodic unit with the phonological word that follows it, but it does not undergo clitic movement. If the subject clitic does not precede the highest head in the extended projection of the verb, it does undergo clitic movement and clusters with object and locative clitics behind the highest head (as in (1)). I investigate the difference between the subject and non-subject clitics throughout this article; I argue that the positioning of the subject clitic does not require it to have a different status than other weak pronouns, contra several other analyses, and that it additionally provides evidence against a prosodic (or mixed syntax–prosody) analysis of clitic placement.

I first establish that the subject markers in Wolof are indeed pronominal clitics and not agreement morphology.

Wolof is an SVO language with a rigid clause-internal word order and several left-peripheral positions for Ā-extracted and left-dislocated elements. All finite clauses in Wolof contain a high projection that hosts complementizer-like morphemes—sentence particles (Dunigan 1994, Martinović 2015, Martinović 2017a, and Martinović 2023; see also Torrence 2005 and Torrence 2013 for a more heterogeneous view of these elements). In the examples that follow, the sentence particles will be underlined. There are two types of finite clauses in Wolof (Martinović 2015 and Martinović 2023): those in which a non-clitic subject is to the left of the sentence particle and is obligatorily doubled by a clitic to the right of the sentence particle, a clause type illustrated by the neutral affirmative sentence in (3a) and the predicate focus sentence in (3b),⁴ and those in which an XP Ā-moves in front of the particle and the subject remains below it, illustrated by the exhaustive identification sentence in (3c) and the relative clause in (3d).

I follow Dunigan 1994, Martinović 2015, Martinović 2017b, and Martinović 2023 in holding that sentence particles all occupy the same position in the clause. They are in complementary distribution with one another, they appear to carry clause-typing information, and a clause must have one in order to have tense and negation (Njie 1978 and Njie 1982). Elements to the left of sentence particles behave as if they were in the left periphery. I therefore identify this position as a (low) C head.⁵

There are a number of syntactic differences between the two clause types in (3). Most strikingly, in the type that does not involve Ā-extraction, exemplified by (3a) and (3b), a non-clitic subject (xale yi ‘the children’) can only be to the left of the sentence particle and is obligatorily doubled by a clause-internal clitic (ñu). The non-clitic subject cannot occur to the right of the sentence particle, regardless of the presence or absence of subject marking:

This makes it appear as if the element encoding the ϕ-features of the subject is agreement, with the non-clitic subject being banned from a post-agreement position due to its need to move to the specifier of Agr, for example. With Ā-extraction of a non-subject, however, the clause-internal subject can either be a clitic, as in (5), or a non-pronominal DP, as in (6). The non-clitic subject cannot co-occur with the clitic to the right of the sentence particle, but it is obligatory to have one of the two.

In local Ā-extraction of a subject pronoun, the pronoun is found to the left of the sentence particle, and a clause-internal subject clitic is not possible (see discussion of (17) below).

The comparison between the two clause types shows that the subject clitic to the right of C should not be treated as agreement (contra Zribi-Hertz & Diagne 2002, Torrence 2005, and Torrence 2013). In Martinović 2015 and Martinović 2023, I propose that clauses like those in (3a) and (3b) involve the dislocation of the non-clitic subject to the left periphery, with the clitic being the true clause-internal subject.⁶ Specifically, I argue that, in clauses that obligatorily dislocate subjects, the clause-internal subject position is not available because the features of C and T are bundled into one head, thus making only one specifier position available. In Ā-movement constructions, on the other hand, C and T head separate projections, thus making one specifier position available for the subject and another for the Ā-moved element. The structure of the two clause types is shown in (7) and (8); I assume this in the rest of the article. (I assume the verb respects the Head Movement Constraint in moving to CT or T, but I omit this from the phrase markers for simplicity.)

The bundling of C and T will be relevant for cliticization in infinitives with overt subjects (see section 5).

In finite matrix clauses, clitics target the position to the immediate right of the sentence particle, clustering in the order subject ≺ object ≺ locative. The examples in (9a–c) show clauses with no Ā-movement and (optional) non-clitic subjects to the left of C, while those in (9d, e) show clauses with Ā-movement.⁷

The non-clitic subject to the left of C in (9a–c) is not focused, and it does not have to be a topic; sentences such as (9a, b) can be uttered out of the blue (Russell 2006; see Martinović 2015 and Martinović 2023 for a detailed analysis of the positions of subjects in the left periphery). Any clause can have a topic; if the subject to the left of C in clauses such as (9a–d) is topicalized, it is followed by a pause (Rialland & Robert 2001). The optional non-topic subject in (9a–c) forms an intonational unit with the material that follows it. The clitics therefore do not occur in a consistent position relative to the left edge of the intonational phrase: in (9a–c), they follow the second phonological word (namely the verb with its affixes and the complementizer if the subject DP is present but the first phonological word if the subject DP is absent); the same variation in clitic placement is found between the Ā-movement constructions in (9d) and (9e). Clitics cannot be initial in finite clauses, which follows straightforwardly from an analysis in which they target the highest functional head in the extended projection of the verb, the sentence particles, which in Wolof I identify as C heads.

The position of the clitics is higher than the structural subject position (Russell 2006), which can be seen when the clause-internal subject is non-pronominal (where this is possible, namely in Ā-movement clauses). Compare (10a), which has a clitic subject, and (10b), with a non-clitic subject.

The examples in (3–10) show that clitics always occupy the same position when a sentence particle is present, independently of the presence or position of any other element in the clause (verb, non-clitic subject, Ā-moved XP). Clitics also occur in structures that do not contain overt C elements, in which case they do not necessarily cluster together. For example, there are clauses that occur in running narrative contexts (storytelling) or in proverbs. They are unembedded, but they cannot contain any inflectional morphology (tense, aspect, negation); they must acquire temporal reference from context or be directly preceded by a temporal adverbial phrase. I refer to them as minimal clauses (Zribi-Hertz & Diagne 2002); an example is given in (11). Pronominal clitics can occur in minimal clauses, as shown in (12).

The fact that the subject clitic can be clause-initial shows that Wolof does not have a prosodic ban on initial clitics; compare, for example, Warlpiri, which Legate 2008 argues involves syntactic clitic placement, except in one instance where post-syntactic reordering is triggered due to the clitics being enclitics and therefore needing to lean on an element to their left. No such requirement appears to be placed on Wolof clitics. The example in (12) additionally shows again that clitics do not occur in a consistent position relative to the left edge of the intonational phrase.

As expected, Wolof clitics form a phonological word with the material that they cliticize to. This can be seen from the application of vowel harmony to the clitic cluster, shown in (13). Wolof has ATR harmony, which spreads from roots to affixes and clitics.

Note that vowel harmony spreads only inside phonological words, and each phonological word contains only one root; for example, the non-clitic subjects to the left of C in (9) do not harmonize with C and clitics. In the remainder of this article, I do not represent vowel harmony but consistently use the –ATR spelling for clitics and affixes.

Additionally, subject clitics have a different form depending on the sentence particle they are adjacent to. For example, the first person singular subject clitic surfaces as a in neutral affirmative clauses, with the sentence particle na, but as ma if there is negation on the verb, in which case the sentence particle is phonologically null; this is shown in (14). The subject clitics and C form portmanteau morphemes in the second person singular and plural, as shown in (15).

This kind of variation in form is present in most persons but not all: for example, the first and third person plural pronouns remain the same in all environments. The phonological form of object and locative clitics does not change depending on the morphosyntactic context. I take this to mean that subject clitics incorporate into C whereas other clitics do not. This is not relevant for the discussion of clitic movement.

Finally, we need to resolve a puzzle related to the surface form of subject pronouns that move to spec-CP, which has complicated some previous analyses of clitics, as discussed in the next sub-section. Cross-linguistically, pronouns surface in their strong form in left-peripheral positions, when they are focused, when they are complements of prepositions, and in coordination. At first glance, Wolof appears to be a counterexample to the generalization that pronouns always surface as strong in focused and/or left-peripheral positions. While an object pronoun that Ā-moves to spec-CP can, as expected, only surface in its strong form, not as a clitic, in the case of subject extraction it appears that the weak version of the pronoun surfaces in spec-CP and not the strong one. This is shown in (16) versus (17).⁸

The asymmetry between subject and object pronouns in extraction is, however, only apparent. The key clue comes from the second person singular and plural, where there is no phonological similarity between weak and strong pronouns (see table 1). In the case of second person subject extraction to spec-CP, the pronoun does not surface in its weak form, nga/ngeen, but as a truncated strong form, ya, in the singular (full strong form is yow) and as the full strong pronoun yeen in the plural:

The generalization that spec-CP is a position in which only strong pronouns can surface therefore holds for Wolof as well; subject pronouns in spec-CP are truncated versions of the strong forms. This is important, as the assumption that subject pronouns in spec-CP are clitics has introduced complications into previous analyses of cliticization in Wolof, as I discuss in section 2.2.

Another argument in favor of this analysis of Ā-moved subject pronouns is provided by pronouns in fragment answers. Fragments have been argued to have full sentential structure, to account for their propositional character. Merchant 2004 proposes that the fragment moves to the specifier of a left-peripheral head, with the TP being elided. Fragment answers in Wolof, which can be followed by the Ā-movement complementizer (l)a, support this. Only strong pronouns can be fragment answers, as shown in (20), regardless of whether or not they are followed by the complementizer (l)a and regardless of their grammatical relation (Martinović 2018). The fragments in (20) can both be answers to either of the questions in (19).

Fragment answers show that the form of the pronoun that surfaces in spec-CP is indeed the strong form, which becomes reduced in subject extraction when the CP layer is followed by overt material. This truncation seems to be a phonological phenomenon, and it occurs not only in spec-CP but also in coordination: a subject pronoun that is the first conjunct is truncated in the same way as subject pronouns in spec-CP. I offer no analysis of the truncation of strong pronouns, as it is not relevant for cliticization.

It is important to note here that the overt subject pronoun in non-finite clauses is not a truncated strong pronoun. In the minimal clause example in (21), for example, the second person singular subject pronoun surfaces in the clitic form nga and not the reduced strong pronoun form ya.

This shows that subject pronouns in initial positions of minimal clauses are indeed clitics.

The discussion thus far has highlighted two important properties of Wolof clitics: (i) they do not surface in a consistent position with respect to the left edge of the clause and/or the intonational phrase, and (ii) a clitic may be initial in the intonational phrase. I shall argue that these properties can be captured by a purely syntactic account of clitic placement. Before going into detail regarding clitic movement in Wolof, I briefly summarize previous analyses of Wolof cliticization.

2.2 Previous analyses

This article affirms the analysis originally proposed in Dunigan 1994. Dunigan argues that clitics move as heads to the highest functional head in the extended projection of the verb, in the sense of Grimshaw 1991. According to Grimshaw, an extended projection is composed of a lexical head, its projection, all of the functional heads that represent features of the lexical head, and the projections of those functional heads. Dunigan proposes that clitics move via long head movement, similarly to what I propose in (25) below, though because she treats subjects in spec-CP of Ā-movement constructions as clitics and not truncated strong pronouns, she is forced to treat subject clitics differently: she proposes that they move as phrases, with some additional assumptions needed to get them to precede object and locative clitics when they follow C. I showed in the previous sub-section that this view of subject pronouns in spec-CP is mistaken, but I will argue that subject clitics can undergo A-movement.

Zribi-Hertz & Diagne 2002 criticizes Dunigan’s purely syntactic account and offers a mixed syntax–prosody analysis. The issue that Zribi-Hertz & Diagne take with Dunigan’s analysis has to do mostly with her treatment of subject pronouns as clitics. They choose to treat subject pronouns as agreement in the clause type exemplified by (3a, b), which I analyze instead as having non-clitic subjects to the left of C and only clitics as clause-internal subjects (i.e., subjects below C); see the discussion that follows (3). Treating subject clitics as agreement in these and only these cases forces Zribi-Hertz & Diagne to divorce the behavior of subjects from other clitics. In all other types of clauses subject clitics cannot be agreement, as they are in complementary distribution with non-pronominal subjects, as shown again in the following Ā-movement example.

Zribi-Hertz & Diagne take this to mean that some finite clauses exhibit subject–verb agreement whereas other ones do not. I argue extensively against this view in Martinović 2015 and Martinović 2023. In a broader Niger–Congo context, it has been independently shown for a variety of languages that subjects in finite declarative clauses are in some left-peripheral/dislocated position, in contrast to subjects in other kinds of clauses (notably, Ā-movement constructions); Wolof is therefore not unusual in obligatorily dislocating the subject in certain kinds of clauses. Given that dislocated elements in Wolof must be co-indexed with a clause-internal pronoun, the occurrence of the clitic is expected.

Zribi-Hertz & Diagne are also troubled by the apparent occurrence of clitics in spec-CP in Ā-movement constructions, which also forced Dunigan to treat subject clitics differently from other clitics; we can dispense with this issue as well, as I have shown that pronominals in spec-CP are truncated strong pronouns, not clitics.

Ultimately, the definition of cliticization that Zribi-Hertz & Diagne offer is almost identical to Dunigan’s but with an added prosodic component:

The prosodic component is needed precisely because Zribi-Hertz & Diagne sometimes treat subject clitics as agreement in a separate head, which first must incorporate into the highest head (in my analysis C), and they argue that this incorporation is post-syntactic. Since other clitics do not precede but follow the subject clitic, their placement must now also be post-syntactic. However, we can see from Zribi-Hertz & Diagne’s definition that the actual clitic placement is syntactic, as it references a syntactic position. The PF then needs to know what the “topmost head” in the extended domain of the verb is, even though the structure has already been linearized and prosodic words have been formed. Ultimately, I believe that it is exactly clitics that show that Zribi-Hertz & Diagne’s view of Wolof clausal structure is not correct.

Another analysis invoking prosodic readjustment is that of Russell 2006. Russell proposes that all clitics in finite clauses move to spec-TP and are then moved as a block to C post-syntactically. Clitics in infinitival clauses disprove this analysis, so I will comment on it in more detail at the end of section 5.

Finally, Torrence 2013, along the lines of Sportiche 1996, assumes that clitics occupy fixed projections in the clause, as in (24), and that the differences in their surface position with respect to other elements in the clause is the result of how high the verb or the verb phrase moves.

Torrence accounts for the different clitic positions via a combination of verb movement, remnant VP movement, and, in some cases, independent head movement of the negation suffix. The different types of movement do not appear to me to be motivated or constrainable in any way. For example, Torrence 2003 assumes that the verb moves as a head in clauses in which negation is not present but that the whole VP moves if it is; Torrence additionally assumes that, just in that case, negation must head-move as well so as to get the correct morpheme ordering. (This is employed to account for the behavior of the optional past tense morpheme oon and is not directly relevant here, but see Martinović 2019 for an alternative analysis.) There is nothing principled that blocks verb movement and forces VP movement in some structures but not in others. It is also not clear exactly where the different clitic positions sit in the clausal spine and what the target for VP movement (labeled XP in (24)) would be.

My goal in this article is to affirm Dunigan’s initial proposal and to additionally show that a uniform analysis can be given for subject and object/locative clitics. We will see moreover that clitics can be a useful tool for identifying the amount of functional structure in a clause. I represent the moved clitics as in (25), right-adjoined to the head that they target.

In the next section, I lay out the details of the analysis of clitic movement in finite clauses.

4.1 Non-finite complements and clitic climbing

The structure and size of infinitives is explored in detail in Martinović 2024, so I will keep the discussion here brief. Wolof non-finite complement clauses are characterized by the following properties. They never contain an overt sentence particle/complementizer, they cannot contain negation, and overt subject pronouns are obligatory when the matrix verb takes a DP object or when the embedded subject is not controlled but are prohibited otherwise. The verb in non-finite clauses is in its bare form; that is, there is no special infinitival morphology. In this section I am concerned with clauses like the one in (32), which have the properties of typical subject control constructions: the embedded subject is unpronounced, control is obligatory, and only sloppy readings are available under ellipsis.

In Martinović 2024, I propose that infinitives come in several sizes. In subject control, complements of typical restructuring predicates (modals, aspectual verbs, ‘try’) are argued to be either vP- or AspP-sized, whereas non-restructuring predicates are argued to take TP-sized complements. The two groups of infinitives line up closely with the well-established distinction between restructuring/tenseless infinitives on the one hand and non-restructuring/tensed infinitives on the other.

All infinitival complements in Wolof, even those that are generally considered to be very small in some languages (e.g., infinitives that Wurmbrand 2001 characterizes as VP-sized in German), can contain the morpheme di. In finite clauses di encodes imperfective aspect, which carries present, habitual, or future meaning; Bochnak & Martinović 2018 treats di as an Asp head, and I assume that it is also an Asp head in infinitives. Evidence for the fact that di involves a head from the inflectional layer lies in its interaction with clitic climbing, shedding further light on properties of cliticization, as I will now show.

Clitics in Wolof obligatorily climb to the matrix C out of complements of typical restructuring verbs, such as modals, aspectual verbs, and implicatives. To illustrate with a modal:

Clitic climbing is optional out of complements of verbs that, cross-linguistically, typically do not restructure, such as desideratives, as illustrated in (34). The object clitic ko can be found either after the embedded verb or attached to the matrix C, following the subject clitic (the optionality of clitic climbing is represented with curly brackets).

In Martinović 2024 I propose that the optionality of clitic climbing is related to the presence of any head from the inflectional layer inside the infinitive: if such a head is absent, clitics obligatorily climb, and if it is present, they optionally climb. Complements of desiderative verbs are always (at least) TP-sized, and therefore clitic climbing is always optional. This is in line with the claim that non-restructuring infinitives contain more structure than restructuring infinitives (e.g., Wurmbrand 2001 et seq. and Grano 2015).

Crucial evidence comes from the complements of otherwise obligatorily restructuring verbs where the presence of the aspectual element di makes clitic climbing optional, just as in non-restructuring infinitives. The example in (35a) shows this for clitic climbing out of the complement of the verb war ‘must,’ and (35b) likewise shows that the clitic can stay in an embedded infinitive that contains the verb jéem ‘try,’ as long as di is present.

I take this to mean that di is an inflectional head. Complements of typical restructuring verbs are therefore either vP-sized (when di is not present), and clitic climbing is then obligatory, or they are AspP-sized (in the presence of di), when clitic climbing becomes optional.

To capture the properties of clitic climbing, I propose that clitics have an additional feature that is checked by a head from the inflectional layer. The introduction of a feature specific to a particular layer of the clause follows the intuition that the clause is divided into “domains” and that each domain is the locus of certain kinds of information/operations. One implementation of this observation is Grohmann 2003’s clausal tripartition into the theta domain (roughly vP/VoiceP), the phi domain (roughly TP, or my “inflectional layer”), and the omega domain (or the left periphery). Grohmann endows the heads in each domain with a “context value.” Following this, I propose that every head in the inflectional domain carries the feature |Φ| and that clitics have u|Φ|.

That clitics have a privileged relationship with the inflectional layer has been, in one way or another, a part of most analyses of clitic movement, though it is usually captured as a relationship with Agr or T (e.g., Kayne 1991) and often related to case licensing. For example, Rezac 2005 proposes that clitics in Czech climb out of infinitives that are VP-sized (according to Wurmbrand 2001), as there are no case licensers present in such a small structure and clitics must be licensed by entering into a relationship with functional categories that they themselves lack, due to their structural deficiencies (along the lines of Cardinaletti & Starke 1999).

All Wolof infinitives can have the aspect morpheme di, which means that no infinitive is only VP-sized; therefore an accusative case assigner would always be present. Even in the presence of di, clitics can climb (though they do not have to). Additionally, it is not clear that v would be involved in the licensing of the locative clitic.

Furthermore, if we take seriously the analyses of future-oriented infinitives, which are argued to be CP/TP-sized or wollP-sized (e.g., Grano 2012, Wurmbrand 2014b, and Grano 2015), then their obligatory reduction to VPs in the instances of clitic climbing should prevent future-oriented interpretations, assuming that those are licensed by additional structure as, for example, Wurmbrand 2014b argues (Wurmbrand proposes that this is what licenses mismatched adverbials in the matrix clause and embedded clause, e.g., Frodo decided today to take the ring to Mordor tomorrow). The expected incompatibility of clitic climbing and future-oriented interpretations is not observed. To preserve the distinction argued to exist between restructuring infinitives, which do not allow future-oriented interpretation with respect to the matrix event (Frodo tried today to leave the Shire (*tomorrow)), and typically non-restructuring ones, which do, I keep the distinction in size between complements of typical restructuring verbs (smaller than TP) and complements of typical non-restructuring verbs (TP) and therefore argue that the relevant feature for clitics is a context feature |Φ| on every head in the inflectional layer, as opposed to, for example, [ϕ] on C or T.

Specifically, I propose that any head from the inflectional layer can establish the required relationship with the clitic, and I assume that this is involved in the licensing of the clitic (I stay agnostic on the question of why such licensing is required). To illustrate, consider the following derivations of obligatory and optional clitic climbing from the infinitival complement of the restructuring verb jéem ‘try’ (from Martinović 2024: 1599–1600).

First, in (36), in the absence of aspect, clitics must climb to the matrix CT. I propose that there are two edge features, one on the highest head of the infinitival complement, here a vP, and one on the matrix CT. Note that I do not claim that the edge feature occurs on every v/phase head, only on the highest head in the extended projection of the verb. Given that I argue that complements of control verbs can be vP-, AspP-, or TP-sized, any of those heads can carry the edge feature. If the clitic in (36) were to climb to the edge of the embedded infinitive, v, this would turn its switch off, but its u|Φ| feature would not be checked. Moving the clitic to the matrix CT results in the checking of u|Φ|, and I argue that this makes clitic climbing obligatory in restructuring constructions without di.

I have here represented the movement to the higher CT as occurring in one fell swoop, seemingly incurring a Minimality violation, as it is skipping over the intervening [ef] on v. I return to the details of how movement to the higher clause proceeds in the next sub-section.

In the presence of di, clitic climbing becomes optional, just as it is optional out of complements of non-restructuring verbs. In Martinović 2024, I take this as evidence for the similarity between the infinitival clause with di in (37) and the infinitival complement of a non-restructuring verb, which I argue to be a full TP. Specifically, I propose that the |Φ| context feature is found on any head in the inflectional layer and therefore any head can check u|Φ| on the clitic. The clitic can therefore surface attached to the highest head in the infinitival complement, Asp, or it can attach to the matrix CT.

Clitic climbing is equally optional with typically non-restructuring subject control verbs, which I argue take a TP-sized complement, and would receive the same analysis as with the restructuring verb jéem ‘try’ in the presence of di. In the next sub-section, I provide an analysis of the optionality of clitic climbing.

4.2 Optionality of clitic climbing

The optionality of clitic climbing introduces different problems depending on whether clitic movement to the higher edge over a lower one is taken to happen in two steps or in one fell swoop.

If clitics first move to the lower edge and then to the higher one, then if they adjoin as heads, it means that they need to excorporate; in section 5, I argue that clitics do have to move as heads, as opposed to undergoing, for example, phrasal movement in the syntax followed by PF adjunction to the highest head. Additionally, if clitics can optionally stay at the lower edge, under the proposed analysis this indicates that their u|Φ| has been checked. The question then arises of how to even motivate their movement to the higher edge: if the edge feature in the matrix clause turns the clitics’ switch on again, we would expect their movement to be obligatory, not optional. And finally, if clitics can either stay at the edge of the infinitival clause or move to the matrix clause, we might expect that they could be split: some clitics might move to the higher edge and others stay at the lower edge. Splitting the clitic cluster has been reported to be possible in some varieties of Serbo-Croatian (Stjepanović 1998 and Stjepanović 1999), but it is not possible in Wolof.

If, on the other hand, clitics skip the lower edge and move directly to the matrix C, they seem to be violating Relativized Minimality (Rizzi 1990). In languages that have optional clitic climbing (e.g., Italian), one account of the optionality has been that in the absence of clitic climbing we are dealing with a non-restructuring configuration (a larger complement out of which clitics cannot climb) whereas in restructuring constructions the complement is very small (e.g., VP), so clitics obligatorily climb. In Wolof, clitic climbing is possible out of structures that are universally argued to be larger than VPs (complements of desiderative verbs), and climbing can occur in the presence of di, indicating that the complements involved cannot be as small as VPs.

I propose that clitics undergo only one instance of movement, either to the lower edge or to the higher edge, and that there is no Minimality violation. I follow the theory of movement developed in Heck 2016, Heck 2022, and Heck 2023. Heck argues that movement consists of two steps:

1. Removal of the element from the tree into a separate workspace (independently proposed in Bobaljik 1995, Uriagereka 1995, Müller 2017, and Müller 2018). Each workspace is associated with a specific feature.

2. Re-merger of the element back into the tree.

If the two operations apply in immediate succession, we cannot identify them. Heck explores a large number of phenomena in which another syntactic process intervenes between removal and re-merger, explaining why a particular structure does not violate either the Strict Cycle Condition (Chomsky 1973) or the Minimal Link Condition (Fanselow 1991, Ferguson 1993, and Chomsky 1995) even though on the surface it appears to do just that. The crucial component of Heck’s theory that makes this possible is the proposal that re-merger can be delayed, or procrastinated,¹⁰ so that removed elements can sit in the workspace while another operation occurs, and be re-merged into the derivation later. In the cases that Heck investigates, the head that triggers the removal of an element into the workspace is also (or its specifier is) the position where the element ultimately re-merges; I propose here that re-merger in the case of cliticization can target a different position.

First, let us investigate, in (38), the option in which the clitic does not climb. When the head with the edge feature [ef] is merged, in this case T, it turns the clitic’s switch on. This triggers the removal of the clitic into a workspace associated with [ef]—step ➀. In step ➋, the clitic is re-merged by being adjoined to T. The clitic’s switch is turned off, and its u|Φ| is checked; I propose that this makes the clitic invisible to subsequent operations, so it cannot be attracted by the edge feature on the matrix CT.

The example in (38) also involves the cliticization of the matrix subject, which proceeds in the same manner—steps ➂ and ➍.

The second option, analyzed in (39), is for the object clitic, which was removed to the workspace by the embedded [ef] on T, not to re-merge with the embedded T but to procrastinate in the workspace and re-merge with the higher head that is also associated with [ef].¹¹ I propose that re-merger can procrastinate because the edge feature is not a feature that needs to be checked, so not re-merging the clitic in the embedded clause does not cause problems on that front; the only feature that does need to be checked is the u|Φ| on the clitic, and there is no requirement for it to be checked in the embedded clause. The edge feature on the matrix CT activates the switch on the matrix subject clitic, a, which I assume is removed into the same workspace where ko has been procrastinating, since their removal is triggered by the same kind of feature. Finally, both clitics, the matrix subject and the embedded object, re-merge with the matrix CT.

Going back to obligatory clitic climbing, movement procrastination allows us to account for that case as well. In (36), there are two edge features, one on the embedded v and one on the matrix CT. I argued that clitic climbing is obligatory because the embedded v does not have |Φ|, so if the clitic were to re-merge with v, its u|Φ| would remain unchecked. Given that I have argued that [ef] and u|Φ| are independent, we have to assume that [ef] on v turns the clitic’s switch on in (36). If the clitic does not procrastinate but merges with v, its u|Φ| is not checked. If it procrastinates and merges with the matrix CT, both its u|Φ| is checked and its switch is turned off. The difference between obligatory and optional clitic climbing is that, in obligatory clitic climbing, only the derivation with procrastination results in a grammatical output whereas, in optional clitic climbing, both the derivation with procrastination and the one without are successful.

In this section, I discussed a property of clitics argued for in Martinović 2024: their need to occupy the inflectional layer of the clause. We also saw how cliticization can be a useful tool in diagnosing clause size. That infinitival complements of control verbs are of different sizes has been independently argued in the literature on control (e.g., Wurmbrand 2001, Wurmbrand 2014a, Wurmbrand 2014b, Grano 2015, Wurmbrand 2015, and Wurmbrand & Lohninger 2023). The patterning of clitic climbing in Wolof lines up with the standard distinction between restructuring and non-restructuring verbs, and what is more, this patterning behaves as expected with changes of clause size. Finally, I proposed a new analysis of the optionality of clitic climbing, relying on Heck 2016, Heck 2022, and Heck 2023’s theory of movement, which allows us to account for the apparent optionality of the movement in a principled manner, without violating Minimality or modifying head adjunction to allow excorporation of clitics (but not other heads in the language).

We now turn to infinitives with overt subjects, which further illuminate the properties of clitic movement and especially the behavior of the subject clitic. I argue that such infinitives offer additional evidence against a prosodic component in clitic placement in Wolof.