I argue that verbal resumption (the occurrence of an additional default verbal element yε meaning ‘do’) in Asante Twi is prosodically conditioned. Following the MATCH theory of syntactic-prosodic constituency correspondence (Selkirk 2011), I propose that phonosyntactic constituency matching requires, at the minimum, avoidance of phonetically empty transferred syntactic structures (i.e., prosodic vacuity). I show that Twi verbal resumption is highly constrained and occurs precisely in those contexts where a prosodically vacuous domain would otherwise be mapped from a fully evacuated syntactic Spell- Out domain. As a measure of last resort, a late default-form insertion of the verb root (the yε-form) occurs to evade prosodic vacuity and ensure a matching correspondence between syntactic and prosodic constituents at PF. Because an additional higher copy of the verb root (i.e., the lexical verb) survives as well, Twi verbal resumption represents an instance of multiple copy Spell-Out. The article thus bears on several issues concerning the syntax-phonology interface, among them the nature of prosodic mapping and the conditions regulating multiple copy realization.
1 Overview: A Curious Distribution
In Asante Twi, there is a striking relationship between clause structure/derivation and verbal resumption (the occurrence of an additional default verbal element). In simple past tense clauses, intransitive verbs (both unergative and unaccusative) are obligatorily followed by nonthematic occurrences of yε, a multipurpose predicate meaning ‘do’, ‘make’, and ‘be’, among other things. The examples in (1a–b) illustrate Twi yε-insertion with both types of intransitive predicates, while those in (2) illustrate the semantic broadness of the yε-form when used as a main verb.
Postverbal yε is limited to the simple past tense. In all other aspects and periphrastic tense constructions,1yε may not follow an intransitive verb.
The polarity of the clause plays a decisive role as well. Under the scope of negation in the past, yε-insertion is blocked. In the affirmative, however, it is obligatory. This contrast is illustrated in (4).
The presence or absence of a surface complement also matters. As illustrated in (5), yε-insertion is systematically unavailable in transitive constructions with unmoved complements, regardless of tense or aspect.
Do-support is typically viewed as a morphologically driven phenomenon that targets the higher regions of clause structure. Twi yε-insertion, however, does not appear to fit this mold: it is neither obviously morphologically motivated (it is a free morpheme that does not serve as a base of affixal attachment) nor structurally high in any clear sense. Despite its formal similarity with the ‘do’-form in the language, I argue in this article that yε-insertion in Twi is less an instance of do-support and more accurately an instance of prosodically conditioned verbal resumption.4 Following the MATCH theory of syntactic-prosodic constituency correspondence (Selkirk 2006, 2009, 2011), I propose that phonosyntactic constituency matching requires, at the minimum, avoidance of phonetically empty transferred syntactic constituents (i.e., prosodic vacuity). I show that Twi verbal resumption is highly constrained and occurs precisely in those contexts where a prosodically vacuous domain would otherwise be mapped from a fully evacuated syntactic domain. As a measure of last resort, a late default-form insertion of a lower copy of the verb root (the yε-form) occurs to evade prosodic vacuity and ensure a matching correspondence between syntactic and prosodic constituents at PF. Because an additional higher copy of the verb root (i.e., the lexical verb) survives as well, Twi verbal resumption represents an instance of multiple copy Spell-Out. This has broad theoretical consequences. As an instance of multiple copy interpretation at PF, Twi yε-insertion provides an opportunity to supplement the growing catalogue of forces known to drive multiple copy realization at the syntax-phonology interface (see Kandybowicz 2008). With respect to prosodic mapping, Twi yε-insertion provides evidence for the existence of a PF constraint banning phonetically null (i.e., vacuous) Spell-Out domains, a constraint that follows logically from the MATCH theory, as I will discuss.
This article has considerable descriptive consequences as well. First, it challenges conceptions of yε in the Twi literature. Existing accounts of yε treat the morpheme as either an additional exponent of past tense (Boadi 1966, Dolphyne 1988), an exponent of completive aspect (Osam 2003), or a contextual allomorph of the past tense morpheme (Ofori 2006). Either way, these approaches view yε as an exceptional instance of verb suffixation5 in a language that is predominantly prefixal.6 Concerning the purported past suffix, Dolphyne (1988:93–94) writes:
This is the only suffix in the verbal forms. . . . There are two different realisations of this suffix depending on whether or not the verb is followed by an object or a complement. . . . Where the verb is not immediately followed by an object or a complement, a Low tone suffix . . . occurs after the stem. In Asante the suffix has two alternative forms, either the high front vowel -i/-e or -yε,7 and in each case the suffix is preceded by a long vowel. . . . Where the verb is immediately followed by an object or a complement . . . the suffix does not occur.
I am unaware of any existing treatment that views yε through the lenses of verbal resumption and late insertion, as conceptualized in this article. My analysis of yε might therefore seem controversial when considered against the backdrop of the existing Twi literature. However, any potential controversy is offset, I believe, by the wealth of correct predictions this perspective brings to bear on the language, predictions that lead to several new discoveries concerning the syntax of Asante Twi. I enumerate these new findings, which I take to be the article’s second descriptive consequence, throughout the following sections.
The article is organized as follows. Section 2 details those aspects of Twi clause structure that are essential for deriving the distribution of yε. It is primarily concerned with the cartography of the Twi middle field, which has yet to receive a formal systematic treatment in the literature. Section 3 returns to the distribution of nonthematic yε and, based on the analysis of Twi clause structure proposed in section 2 and a closer look at the facts, offers a descriptive generalization covering a wide range of data, both previously established and newly discovered: namely, the Spell-Out domain of v0 must have content at PF; otherwise, yε is inserted to supply that content. Section 4 derives this descriptive generalization from deeper principles concerning prosodic mapping. Section 5 concludes the article with a summary and brief closing remarks.
2 Twi Clause Structure
2.1 The Syntax of T0
A number of important insights into Twi clause structure accompany the examination of T0Past. The head has two realizations depending on whether or not aspect is encoded: a null variant occurs when the clause contains no marking of aspect and a High-tone-bearing na variant surfaces exclusively in conjunction with aspect. These two expressions of T0Past are illustrated in (7).
Not all linguists studying Twi make this two-way distinction. Osam (2003), for instance, treats na in (7b) as an adverb rather than a true tense morpheme, pointing out that it is homophonous with the form meaning ‘then’ that surfaces in conditionals. Telling evidence against Osam’s analysis of na, however, comes from extraction facts. Unlike true adverbials in the language, which may be focused (see (8b)), sentence-initial na resists movement8 (see (8d)), as expected of a true tense morpheme.9
εnora Kofi kɔɔ dwaso.
yesterday Kofi go.PST market
‘Yesterday Kofi went to the market.’
εnora na Kofi kɔɔ dwaso
yesterday FOC Kofi go.PST market
‘It was YESTERDAY that Kofi went to the market.’
Na Kofi a-kɔ dwaso.
PST Kofi PRF-go market
‘Kofi had gone to the market.’
*Na na Kofi a-kɔ dwaso.
PSTFOC Kofi PRF-go market
One potential explanation of (8d)’s ungrammaticality that would be consistent with Osam’s claim would involve an appeal to haplology. Along these lines, one could maintain that na is in fact an adverbial expression, but because it is homophonous with the focus marker (though see footnote 9), it may not be focused because doing so would yield identical adjacent na sequences at PF. Evidence against such an appeal is furnished by examples like (9), which illustrate that the locative object in (8c) is capable of undergoing focus movement, yielding a grammatical surface na na string.
Dwaso na na Kofi a-kɔ.
market FOCPST Kofi PRF-go
‘It was THE MARKET that Kofi had gone to.’
Although segmentally null, one important phonological reflex of T0Past can be detected. This reflex sheds light on the structure of the Twi clause as well as its derivation. When the null variant occurs and no overt head intervenes between T0 and V0, the predicate’s final segment (typically, a vowel) is lengthened/copied (see (10a)). However, when an overt head (such as negation10 (see (10b)) or aspect (see (10c))) intervenes, the verb’s final segment is obligatorily short. (See Dolphyne 1988 and Paster 2010 for more details.)
These facts suggest that Twi verbs raise to T0 unless blocked by an overt/contentful head (see Aboh 2004 on Gbe) and that when the verb occupies T0Past, it is spelled out with an additional mora (see Kobele and Torrence 2006). In this way, segment-final lengthening on a verb can serve as a diagnostic for V0-to-T0 movement. Unfortunately, the standard tests for V0-to-T0 raising are not applicable in Twi because the language lacks verb phrase–initial adverbial modifiers (Saah 2004:70) and does not tolerate quantifier float in vP-internal subject positions (more on the latter in section 2.3).
Word order facts in sentences with overt T0Past, in which an aspect marker obligatorily cooccurs with the past tense morpheme, also shed considerable light on Twi clause structure. When the two cooccur, the subject must follow T0 and precede aspect. Furthermore, because the verb fails to raise to T0, it is spelled out with an obligatorily short vowel. And as first exemplified by the data in (3b,d), when T0Past and aspect cooccur, yε may not accompany the verb. This is illustrated first in sentences (11a–c) and then again in (11d–f ) with a different predicate and aspect marker.
Although these word order facts might suggest that na has raised from T0 to C0, cooccurrence facts in embedded clauses tell otherwise. Example (12) shows that na is not in complementary distribution with C0 and thus that the two items do not compete for the same syntactic position.
Yaw kaa sε na Kofi re-sa.
Yaw say.PSTCOMPPST Kofi PROG-dance
‘Yaw said that Kofi was dancing.’
What, then, is the proper analysis of the na-construction in terms of the positions occupied by the tense marker and the subject? As it turns out, answering this question unlocks the structure of the Twi middle field.
2.2 The Structure of the Na-Construction
The acceptability of (12) reveals that the presubject position of na is not C0. Eliminating this analytical possibility, two analyses of na would seem equally reasonable at this point. One would involve locating the item in a low left-peripheral head above T0—for instance, FinP (Rizzi 1997). On this approach, one could remain agnostic about whether na was base-generated in this position or moved there from T0. Crucially, na would be hierarchically superior to the subject in Spec,T0, correctly deriving the word order facts in (11). Call this Analysis A, illustrated in (13).
A contending analysis, call it Analysis B, would generate na directly in T0 and account for the linear order of the tense marker and subject by locating the latter vP-internally. This is illustrated in (14).
On Analysis B, movement of the subject to Spec,T0 would be suspended, presumably because of an exceptional absence of EPP features on na.11
2.3 Deciding between Analyses A and B
The crucial difference between Analyses A and B concerns the position of the subject. On Analysis A, the subject has raised, occupying Spec,T0. On Analysis B, the subject is unmoved, occupying its base Merge position, Spec,v0. The distribution of floating quantifiers in the language provides a means for locating the syntactic position of the subject in the na-construction.
In Asante Twi, a quantifier like nkoara ‘only’ follows the DP it quantifies (15a). When its associate DP is displaced, a quantifier may be either pied-piped (15b) or stranded (15c). However, when stranded, the quantifier must be accompanied by a preceding strong resumptive pronoun agreeing in φ-features with the displaced associate.
Kofi a-bɔ [Ama nkoara].
Kofi PRF-kick Ama only
‘Kofi has kicked only Ama.’
[Ama nkoara] na Kofi a-bɔ _____.
Ama only FOC Kofi PRF-kick
‘It was ONLY AMA that Kofi has kicked.’
Ama na Kofi a-bɔ *(ɔno) nkoara.
Ama FOC Kofi PRF-kick 3SG only
‘It was only AMA that Kofi has kicked.’
The data in (15) illustrate that quantifier float is possible in object position. Similar facts reveal that quantifier stranding is also possible in the surface subject position, Spec,T0.
[Kofi nkoara] a-bɔ Ama.
Kofi only PRF-kick Ama
‘Only Kofi has kicked Ama.’
[Kofi nkoara] na *(w)-a-bɔ Ama.
Kofi only FOC 3SG-PRF-kick Ama
‘It was ONLY KOFI that has kicked Ama.’
Kofi na *(ɔno) nkoara a-bɔ Ama.
Kofi FOC 3SG only PRF-kick Ama
‘It was only KOFI that has kicked Ama.’
Example (16a) shows once again that quantifiers follow their associates. (16b) introduces a new and important wrinkle. A quantifier may be pied-piped, as before, but when the [DP + Q] constituent is extracted from the surface subject position, a weak resumptive pronoun must mark the launching site of movement, a pattern typical of subject extraction in the language more generally. (In the case of (16b), the weak third person singular pronominal /ɔ-/ surfaces as [w-] for phonological reasons relating to hiatus resolution.) Sentence (16c) illustrates the previous pattern observed in object position quantifier float: when stranded, the quantifier must be preceded by resumptive :no, the strong form of the third person singular pronoun. Comparable facts obtain in the progressive aspect. The data in (17) recapitulate the pattern in (16), illustrating once more that quantifier float in Asante Twi is tolerated in Spec,T0, the surface subject position.
[Kofi nkoara] rε-bɔ Ama.
Kofi only PROG-kick Ama
‘Only Kofi is kicking Ama.’
[Kofi nkoara] na ɔ-rε-b: Ama.
Kofi only FOC 3SG-PROG-kick Ama
‘It is ONLY KOFI that is kicking Ama.’
Kofi na *(ɔno) nkoara rε-bɔ Ama.
Kofi FOC 3SG only PROG-kick Ama
‘It is only KOFI that is kicking Ama.’
An interesting twist emerges, however, when we consider quantifier float in vP-internal subject positions. Despite the availability of quantifier float in the surface subject position, Twi prohibits quantifier stranding in Spec,v0, as (18) illustrates.
*Kofi bɔɔ [vP ɔno nkoara Ama].
Kofi kick.PST 3SG only Ama
‘Only Kofi kicked Ama.’
*Kofi saa [vP ɔno nkoara yε].
Kofi dance.PST 3SG only yε
‘Only Kofi danced.’
To recap, quantifier stranding is attested in Asante Twi, but it is restricted from occurring in vPinternal subject position. This fact can now be pressed into service in choosing between Analyses A and B. To do so, we will consider the predictions made by both analyses for quantifier float in the na-construction.
Although they differ in where they locate the subject in the na-construction, Analyses A and B both correctly predict the unavailability of subject-oriented quantifier stranding in na-clauses.
Na [Kofi nkoara] a-bɔ Ama.
PST Kofi only PRF-kick Ama
‘Only Kofi had kicked Ama.’
*Na Kofi ɔno nkoara a-bɔ Ama.
PST Kofi 3SG only PRF-kick Ama
‘Only Kofi had kicked Ama.’
*Na Kofi a- ɔno nkoara bɔ Ama.
PST Kofi PRF 3SG only kick Ama
‘Only Kofi had kicked Ama.’
The absence of the strong resumptive element ɔno reveals that the quantifier has not been stranded in (19a). By contrast, (19b–c) involve subject-oriented quantifier float and are therefore ungrammatical. On Analysis A (see (13)), where na is located in a left-peripheral head just above the moved subject in Spec,T0, the ungrammaticality of (19b–c) follows from the independently established restriction on floating quantifiers in Spec,v0. On Analysis B (see (14)), where na is located in T0 and the subject is in situ in Spec,v0, the unavailability of quantifier stranding follows from the fact that the subject is unmoved in the construction. Without subject movement, the syntactic context for pronominal resumption, which licenses the stranded quantifier, is bled. In this way, both analyses successfully predict the impossibility of subject-oriented quantifier float in neutral na-constructions. However, the two approaches make different predictions with respect to subjectoriented quantifier stranding in na-constructions with focus.
The pivotal question is whether quantified subjects may be focused in the na-construction, stranding their associate quantifiers. Because the launching site of this movement would be Spec,T0 under Analysis A (see (13)), a position independently shown to host quantifier float in the language (see (16c), (17c)), the analysis would predict the possibility of subject-oriented quantifier stranding in subject-focused na-constructions. By contrast, the launching site of subject focus movement under Analysis B (i.e., Spec,v0; see (14)) is not a position independently known to support stranded quantifiers. Thus, Analysis B makes the opposite prediction: subject-oriented quantifier stranding should be impossible in na-constructions with focused subjects. The data in (20) show that Analysis B’s prediction is borne out. Subject-oriented quantifier float is not possible when the subject has been focused in a na-construction (20c).
Na [Kofi nkoara] a-bɔ Ama.
PST Kofi only PRF-kick Ama
‘Only Kofi had kicked Ama.’
[Kofi nkoara] na na w-a-bɔ Ama.
Kofi only FOCPST 3SG-PRF-kick Ama
‘It was ONLY KOFI that had kicked Ama.’
*Kofi na na ɔno nkoara a-bɔ Ama.
Kofi FOCPST 3SG only PRF-kick Ama
‘It was only KOFI that had kicked Ama.’
We thus have motivation to reject Analysis A in favor of Analysis B. The adopted structural analysis of the tense marker and subject in the na-construction under Analysis B is presented in (21) (repeated from (14)).
2.4 Structural Implications of Analysis B
The reason the na-construction unlocks the clause structure of Twi is that it affords us a rare opportunity to observe the subject in a low position in the middle field. Recall that in the naconstruction, subjects obligatorily precede aspect markers (see (11a) vs. (11c) and (11d) vs. (11f )). If the subject is vP-internal when na spells out T0Past, as in Analysis B, then it follows that aspect must be expressed fairly low in the Twi clausal hierarchy (i.e., below v0).12 This low vP-internal aspect position corresponds to ‘‘inner aspect’’ (Travis 1991, 2010, Pearson 2001),13 an independently motivated functional projection within the verb phrase shell structure. Given these considerations, then, the gross structure in (22) for the Twi middle field emerges.
Building on the current analysis, we can elaborate on the structure in (22) by considering the hierarchical position of negation. We have seen that negation follows the subject and blocks segment-final lengthening in simple past verbs (see (10b)). This is consistent with an analysis in which Neg0 is situated somewhere below T0 (contra Kobele and Torrence 2006), blocking head movement of the verb into T0. The data in (23a–c), in which the preverbal negative morpheme n- obligatorily follows both the subject in Spec,v0 and the aspect marker in the na-construction, suggests that negation is expressed fairly low in the language—below inner aspect in the verb phrase. While this hierarchical positioning of negation may seem odd from a Eurocentric perspective, low vP-internal analyses of negation have recently been motivated in a variety of languages by different researchers (see Kamali and Samuels 2008, Kahnemuyipour and Kornfilt 2011, Su 2012). The low hierarchical positioning of negation is not limited to the na-construction, however. Whenever negation and overt aspect cooccur, the former always follows the latter, as illustrated in (23d–e).
Na Kofi re-n-sa.
PST Kofi PROG-NEG-dance
‘Kofi was not dancing.’
*Na Kofi n-re-sa.
PST Kofi NEG-PROG-dance
*Na n- Kofi re-sa.
PSTNEG Kofi PROG-dance
‘Kofi did not dance.’
Putting it all together, then, the structure and derivation of the Twi middle field that I have argued for is presented in (24)–(26). The structure in (24b) represents a negative monotransitive sentence in the past tense, a construction formed by way of a null past tense morpheme in T0 and an obligatory perfect marker a- in Asp0. Because the two heads immediately dominating the verb root are filled by Neg0 and Asp0, verb movement is blocked, preventing the verb from reaching T0 and subsequently ensuring that it surfaces vP-internally with a short vowel.
By way of contrast, the tree in (25b) represents the structure of an affirmative monotransitive construction in the simple past tense. I assume that affirmative structures fail to project NegP and thus, because no overt tense or aspect morphemes have been merged in this case, the verb can successfully raise to T0, feeding segment-final vowel lengthening at PF.
And to conclude, the structure of a negative monotransitive construction in the past perfect is given in (26b). Here we find all expressions surfacing in their base-generated positions: the subject appears vP-internally below the tense marker owing to absence of EPP features on T0na, and the overt exponents of T0Past, Asp0Perf, and negation block the verb from raising out of √P. Consequently, the verb fails to escape vP and is phonetically realized with a short vowel.
3 A Closer Look at the Distribution of yε
3.1 Conceptualizing the Basic Facts
The syntactic considerations of the previous section give way to a more focused characterization of the yε-insertion facts. Cross-referencing the data in (1)–(6) with the structural analyses sketched in (24)–(26), a generalization emerges concerning the distribution of yε. Assuming the architecture of the derivation-by-phase framework (Chomsky 2000, 2001), yε-insertion is obligatory whenever AspP, the Spell-Out domain (SOD) of the v0 phase head, would otherwise be empty at PF. Put another way, yε-insertion into an AspP-internal head circumvents the violation of *[SOD ∅], a PF constraint prohibiting phonetically vacuous Spell-Out domains that I propose in this article.14 The basic facts presented in section 1 serve to reinforce this view. In the affirmative simple past (see (27) (= (1)), intransitive verb roots raise to T0, creating vacant AspP structures if all chain links apart from the chain heads are deleted at PF (the typical outcome of chain resolution (Nunes 1999, 2004)). Yε-resumption in this case provides a work-around.
By contrast, when verb movement is blocked entirely, such as when an inner aspect head is merged (see (28a–e), = (3a–e)) or when perfect-accompanying negation is merged (see (28f ), = (4a)), the verb is phonetically realized in its base-generated position within AspP, obviating the insertion of yε.
With verbal complements, whether DPs or CPs (see (29), = (5)), similar considerations are at play. Because these complements independently provide PF content to AspP regardless of whether V0-to-T0 movement takes place, the need for yε-insertion is obviated.
Kofi bɔɔ (*yε) Ama (*yε).
Kofi kick.PSTyε Ama yε
‘Kofi kicked Ama.’
Kofi rε-bɔ (*yε) Ama (*yε).
Kofi PROG-kick yε Ama yε
‘Kofi is kicking Ama.’
Yaw kaa (*yε) sε Kofi bɔɔ (*yε) Ama (*yε).
Yaw say.PSTyεCOMP Kofi kick.PSTyε Ama yε
‘Yaw said that Kofi kicked Ama.’
Yaw a-ka (*yε) sε Kofi taa-bɔ (*yε) Ama (*yε).
Yaw PRF-say yεCOMP Kofi HAB-kick yε Ama yε
‘Yaw has said that Kofi kicks Ama (habitually).’
Kofi saa (*yε) ntεm (*yε).
Kofi dance.PSTyε quickly yε
‘Kofi danced quickly.’
Kofi saa *(yε) ampa.
Kofi dance.PSTyε truly
‘It’s truly the case that Kofi danced.’
Low manner adverbs like ntεm ‘quickly’ in (30a) are presumably AspP-internal elements (Ernst 2002). As such, they supply PF content to AspP at Spell-Out, rendering yε-insertion unnecessary. Higher-attaching speaker-oriented adverbs like ampa ‘truly’ in (30b), on the other hand, take scope over propositions and thus attach well outside the AspP constituent. Since they are AspPexternal elements, they do not provide phonetic content to the Spell-Out domain in any way. And as such, they fail to obviate yε-insertion.
This particular way of conceptualizing Twi verbal resumption thus seems initially promising. In addition to the considerations of the previous paragraph, a wide range of empirical evidence supports the generalization that yε-insertion is a strategy to repair vacuous Spell-Out domains at PF. I turn now to this evidence.
3.2 Cases Where Linear Order Is Irrelevant
A simple alternative to the generalization expressed above can be found in the existing Twi literature. Osam (2003) treats yε as a contextual allomorph of the completive suffix that surfaces if and only if an intransitive past tense–inflected verb form appears clause-finally. That is, on this approach yε-insertion is linked directly to linear order. Osam’s analysis successfully accounts for yε-insertion following past-inflected intransitives. It also derives the obviation of yε-insertion in intransitive aspectual constructions because in these cases the verb root’s failure to raise to T0 prevents it from being inflected for past tense, despite occurring sentence-finally. Osam’s approach can also account for the obviation of yε-insertion in negative intransitive constructions in a similar fashion, since here too the predicate fails to reach T0 and thus does not inflect for past tense. Because Twi is an SVO language, Osam derives the obviation of yε-insertion by surface objects/complements. Osam’s approach also correctly predicts that postverbal modifiers obviate yε-insertion in intransitive constructions, since their inclusion would effectively disrupt the linear order that would otherwise feed yε-insertion.
Kofi saa (*yε) komm (*yε).
Kofi dance.PSTyε quietly yε
‘Kofi danced quietly.’
However, Osam’s analysis breaks down in the face of sentence-final AspP-external modifiers. The example in (32) (= (6b)) shows that clause-final modifiers attaching higher than AspP fail to obviate yε-insertion, a consideration unnoticed by Osam.
Kofi saa *(yε) ampa.
Kofi dance.PSTyε truly
‘It’s truly the case that Kofi danced.’
Osam’s linear analysis wrongly predicts the impossibility of yε-insertion in cases like (32), where a sentence-final adverb removes the context for insertion. When postverbal adverbs take scope over entire propositions and not just AspPs, however, yε-insertion is obligatory even though the verb is nonfinal in the clause. This shows that Twi verbal resumption is sensitive to hierarchical structure and not linear order, as Osam claims. A revision of Osam’s generalization in line with the observed data would be to say that yε is inserted in the completive aspect when it occurs AspP-finally.
There is further evidence that linear order is irrelevant, all based on data previously unreported in the literature. Clause-final interrogative particles fail to obviate yε-insertion in intransitive constructions. Example (33) shows that yε-insertion is obligatory even though the past tense– inflected predicate is not clause-final.
Kwame wuu *(yε) anaa?
‘Did Kwame die?’
The *[SOD ∅] theory of yε-insertion advanced above easily accommodates this fact. If anaa is a left-peripheral interrogative operator (as seems reasonable), then the AspP in (33) would surface as a vacant domain at PF if not for the insertion of yε.
Markers of coordination also fail to obviate yε-insertion. Example (34) shows that yε is obligatory in the initial conjunct, even though its intransitive predicate does not appear in an otherwise sentence-final position.
Kwaku dii *(yε) na :-daa *(yε).
Kwaku eat.PSTyε and 3SG-sleep.PSTyε
‘Kwaku ate and slept.’
(35) [CoordP [TP Kwaku dii yε] [Coord′ na [TP ɔ-daa yε]]]
Under such an analysis, the verbs of both TP conjuncts would successfully raise to T0Past, leaving behind potentially vacant AspPs. In the usual manner, yε-resumption at PF improves an otherwise defective structure.
Similarly, particles following relative clauses do not obviate the insertion of yε, even though they create a linear context where the predicate is no longer clause-final.
ɔbarima no na ɔ-daa *(yε) no
boy the COMP 3SG-sleep.PSTyεCL.DET
‘the boy who slept’
If we understand the particle no following the relative clause in (36) to be a clausal determiner (as seems reasonable given its syntactic position and homophony with the definite determiner), then it occupies an AspP-external position. The obligatory insertion of yε in this case is thus entirely expected. Because the relative clause–internal intransitive verb has raised to T0, the relative clause–internal AspP will be vacant at PF unless yε-insertion applies.
These considerations strengthen the case for conceptualizing Twi yε-insertion in terms of hierarchical structure, as opposed to linear order. The data in (32)–(36) exemplify instances where, despite appearances to the contrary, a vacant AspP structure correlates with verbal resumption. In (32), the postverbal propositional modifier ampa attaches higher than AspP. In (33), the interrogative particle anaa takes scope well over AspP in C0. In (34), the coordinator na operates outside the AspP domain to conjoin two clauses. And in (36), the clausal determiner no takes scope over the embedded clause containing AspP. In all four cases, the verb roots have raised to T0 and the postverbal functional material exists outside AspP. In all four cases, AspP has no PF content independent of yε.
3.3 Cases Where Movement Plays a Role
Another novel finding concerning the distribution of yε is that its insertion is not actually limited to intransitive clauses in the simple past, as Osam (2003) proposes. The hypothesis that yε- insertion in Twi repairs an otherwise illicit AspP structure at PF (i.e., one with no phonetic content) is strengthened by cases involving both V0-to-T0 movement and object extraction in transitive constructions. In these cases, the evacuation of all AspP-internal material creates outputs that are potentially problematic at PF under the *[SOD ∅] account. As predicted, yε-insertion is obligatory when both verb raising to T0 and object extraction occur and it is obviated either when the object appears in situ or when the verb fails to raise to T0. The data in (37) illustrate these facts. (37a) shows that yε-insertion is unavailable when an object wh-expression remains in situ, despite the verb’s having raised to T0. The very same sentence requires yε-insertion, however, when the whelement undergoes movement to the left periphery (37b). In an aspectual construction other than the simple past, where verb movement to T0 does not proceed and as a consequence the predicate is spelled out AspP-internally, the surface position of the object has no bearing on yε-insertion. This is exemplified by the present progressive sentences in (37c–d). In these cases, yε-insertion is blocked regardless of whether or not the object undergoes extraction.
Ama dii (*yε) εdeεn (*yε)?
Ama eat.PSTyε what yε
‘What did Ama eat?’
εdeεn na Ama dii _____ *(yε)?
what FOC Ama eat.PST
‘What did Ama eat?’
Ama re-di (*yε) εdeεn (*yε)?
Ama PROG-eat yε what yε
‘What is Ama eating?’
εdeεn na Ama re-di _____ (*yε)?
what FOC Ama PROG-eat yε
‘What is Ama eating?’
These facts are fully general, as illustrated by the (contrastive) non-wh-focus movement data in (38) and (39). Once again, unmoved objects in the simple past block yε-insertion (see (38a)) because they provide content to the inner AspP at PF, rendering insertion unnecessary. When they are moved in conjunction with V0-to-T0 raising (see (38b)), however, yε-resumption once again correlates with an otherwise vacant AspP structure at PF. And as before, object extraction feeds yε-insertion only in the case of the simple past tense (compare (38b) with (38d)) because it is only in the simple past tense that the verb root escapes vP.
Ama nomm (*yε) nsuo (*yε).
Ama drink.PSTyε water yε
‘Ama drank water.’
Nsuo na Ama nomm _____ *(yε).
water FOC Ama drink.PSTyε
‘It is WATER that Ama drank.’
Ama re-nom (*yε) nsuo (*yε).
Ama PROG-drink yε water yε
‘Ama is drinking water.’
Nsuo na Ama re-nom _____ (*yε).
water FOC Ama PROG-drink yε
‘It is WATER that Ama is drinking.’
Finally, consider the consequences of the interaction between V0-to-T0 raising and object extraction in ditransitive structures. In Twi, it is not possible to focus multiple XPs in a single clause, but it is possible to focus either the goal or the theme in a double object construction. As the proposed analysis predicts, neither derivation should feed yε-insertion because the unmoved object (whether goal or theme) will remain AspP-locked, providing content to the AspP Spell-Out domain of v0 at PF. This prediction is borne out, as illustrated in (39).
Kofi brεε Ama sika.
Kofi bring.PST Ama money
‘Kofi brought Ama money.’
Ama na Kofi brεε (*yε) sika (*yε).
Ama FOC Kofi bring.PSTyε money yε
‘It was AMA who Kofi brought money to.’
Sika na Kofi brεε (*yε) Ama (*yε).
money FOC Kofi bring.PSTyε Ama yε
‘It was MONEY that Kofi brought Ama.’
3.4 Cases Involving Other AspP-Internal Items
Patterns involving modifiers, null object pronouns, and bimorphemic verbs further suggest the influence of a PF constraint banning phonetically empty AspP structures in Asante Twi. First, consider the consequences of the locus of modification. As previously discussed, when the verb root raises to T0Past, only low AspP-internal modifiers obviate yε-insertion. As illustrated in (40), hierarchically low manner adverbials like ntεm ‘quickly’ and komm ‘quietly’ remove the context for yε-insertion (40a), while higher-attaching adjuncts such as temporal modifiers (e.g., εnora ‘yesterday’ (40b)) and speech act–oriented expressions (e.g., ampa ‘truly’ (40c)) do not.15
Further evidence that Twi verbal resumption is driven by avoidance of prosodically vacuous AspP Spell-Out domains is found in pronominal object constructions. Third person singular inanimate object pronouns in the language are phonetically null, as shown in (41).
Unlike overt objects (see (42c)), however, null pronominals do not obviate yε-insertion when the verb raises to T0 in the simple past, as shown in (42a–b). This once again suggests that the operative condition on yε-insertion is PF-centric, rather than syntactic.
Supporting evidence can also be found in the domain of two-part verbs. Twi has a rich inventory of complex predicates, among them inherent complement verbs and split verbs. Some examples appear in (43).
In the simple past, only the initial piece of the complex predicate is inflected/lengthened, suggesting that it alone vacates the verbal domain. This is illustrated in (44) for the split verb variety.
This fact entails that *[SOD ∅] will always be satisfied in two-part verb constructions regardless of tense or aspect, making the prediction that yε-insertion should be systematically obviated in bimorphemic intransitive verb constructions. As predicted, when used intransitively in the simple past, two-part verbs neither require nor trigger yε-insertion because the second morpheme provides content to the AspP domain. The examples in (45) illustrate.
4 Deriving the Distribution of yε from Prosodic Mapping
Assuming the architecture of the derivation-by-phase framework, I have argued that the distribution of yε is systematic and predictable if we also assume the existence of an active PF constraint banning prosodically vacuous Spell-Out domains. Twi yε-insertion is tied exclusively to the affirmative simple past because it is in this construction alone that the lexical verb both escapes and is pronounced outside AspP, the Spell-Out domain of Twi’s v0 phase. The canonical PF outputs of intransitive, object extraction, and null object constructions violate this constraint when the verb raises to T0 (i.e., in the affirmative simple past), unless yε is inserted inside AspP to provide content to the Spell-Out domain at PF. The precise hierarchical position targeted by yε-insertion is difficult to ascertain on purely empirical grounds. Assuming the clause structure argued for in section 2 and reprised in (46), there are two eligible head positions that could potentially serve as the locus of verbal resumption and thus provide content to the lower Spell-Out domain when the verb has raised to T0Past: Asp0 and √0. These heads host lower copies of the root morpheme, one of which I claim is spelled out as the impoverished/default (in the sense of Bonet 1991 and Halle 1997) verb form yε, assuming the Distributed Morphology hypothesis of competitive late insertion (Halle and Marantz 1993, Marantz 1994, Halle 1997).
The choice of which head hosts yε cannot be made on the basis of word order alone, however, since in all instances of verbal resumption, no item intervenes between the past-inflected verb in T0 and yε. Thus, inserting yε into either of these two heads will both satisfy the language’s antivacuity constraint and yield the attested word order facts. Yε-insertion has the character of a last resort repair strategy in the language: it is obviated when the verb, an object, an aspect marker, a negative morpheme, or a modifier surfaces AspP-internally to supply prosodic content to the transferred structure.
Given its syntactic status as a Spell-Out domain, it follows that AspP constitutes a major prosodic domain (i.e., phonological phrase) in Twi (e.g., Dobashi 2003, Selkirk 2006, Adger 2007, Ishihara 2007, Kratzer and Selkirk 2007, Kahnemuyipour 2009, Revithiadou and Spyropoulos 2009, Sato 2009). The question that arises naturally at this point, then, is why major prosodic domains are policed for prosodic vacuity. In this section, I speculate on the motivation for Twi’s antivacuity constraint, seeking to derive the output condition from deeper principles concerning prosodic mapping.
I propose that Selkirk’s (2006, 2009, 2011)MATCH theory of syntactic and prosodic constituency correspondence provides the insight needed to understand why an antivacuity constraint emerges in the grammar of Twi. MATCH theory articulates the view that the constituent structures of syntax and phonology correspond to or mirror each other. In informal terms, this entails that the edges or boundaries of syntactic constituents (be they words, phrases, or clauses) must correspond with the edges of their related prosodic counterparts ( prosodic words, phonological phrases, and intonational phrases, respectively). For the purposes of this article, the match relation between a syntactic XP and a phonological phrase (φ) is relevant.
MATCH PHRASE (Selkirk 2011): MATCH(XP, φ)
A phrase in syntactic constituent structure must be matched by a corresponding prosodic constituent φ in the phonological representation. The left and right edges of syntactic constituent XP must correspond to the left and right edges of prosodic constituent φ in the output phonological representation.
Because prosodic domain formation requires at the bare minimum overt prosodic content (i.e., the absence of phonetic content precludes the formation of an accompanying prosodic domain), a necessary condition for syntactic-prosodic constituency correspondence is the avoidance of prosodic vacuity. In other words, for a given syntactic XP and its corresponding phonological phrase to match in the way prescribed by (47), the spell-out of XP must be phonetically contentful. Otherwise, XP will not be prosodically mapped and a syntactic constituent will have no corresponding prosodic counterpart at PF. In the case of the Twi v0 phase, a phonetically empty AspP structure will fail to trigger prosodic domain formation and subsequently, without the construction of its corresponding phonological phrase, MATCH(AspP, φ) will be violated. To remedy this state of affairs, verbal resumption into either Asp0 or √0 takes place, providing the prosodic content that feeds φ formation, thus facilitating syntactic-prosodic constituency correspondence upon spell-out of the lower phase. Note that MATCH theory on its own makes no claim about prosodic vacuity avoidance. That is, there is nothing in the theory that explicitly says, ‘‘No prosodic constituent should be empty.’’ I take the ban on null transferred domains to be a consequence of prosodic mapping, derived as a corollary of MATCH PHRASE under a strong reading of (47). In this way, the proposal connects with other treatments of edge/domain emptiness in the literature (e.g., An 2007a,b, Kandybowicz 2009) that seek to ground vacuity avoidance in terms of prosodic mapping and alignment.
In Optimality Theory terms, Twi’s MATCH PHRASE constraint must be either undominated or at least fairly high-ranked; otherwise, constraints penalizing the insertion of material not present in the input (i.e., DEP) would block the insertion of yε and consequently yield prosodically empty AspP structures. Despite its status in Twi, MATCH PHRASE (or any of the MATCH family of constraints, for that matter) is not universally undominated. Crosslinguistically, prosodically empty Spell-Out domains are not as strictly policed as they are in Twi. In French, for instance, a language with independently motivated V0-to-T0 movement (Pollock 1989), intransitive verbs may freely raise out of the verbal domain without the accompanying insertion of default prosodic material to facilitate syntactic-prosodic constituency matching. This entails that MATCH PHRASE and its corollary *[SOD ∅] are violable constraints. In languages like Twi, the constraints are absolute, whereas in languages like French they may be sacrificed in order to comply with higher-ranked prosodic markedness requirements. Crosslinguistic variation with respect to prosodic vacuity, then, follows from the differential ranking of MATCH PHRASE in relation to other active languagespecific prosodic markedness constraints, as dictated by MATCH theory and Optimality Theory more generally.
5 Concluding Remarks
This article makes a case for the avoidance of vacuity in prosodic mapping. Assuming the correspondence of syntactic and prosodic constituent structure, as postulated under MATCH theory, it follows that matching requires, at the minimum, avoidance of phonetically empty transferred syntactic structures. Otherwise, a narrow syntactic constituent will fail to be prosodically mapped and will consequently have no corresponding prosodic counterpart at PF in violation of the MATCH PHRASE constraint (47). If, in a given language, avoidance of prosodic vacuity is a low PF priority (as in French, for example), syntactically vacant transferred XPs will be spelled out with no accompanying prosodic repairs or modifications. But if a language prioritizes vacuity avoidance at PF, then a syntactically evacuated Spell-Out domain will be prosodically vulnerable, militating the invocation of some language-specific PF repair strategy.
This issue comes to a head in Asante Twi, where in certain simple past tense constructions an otherwise empty AspP structure is repaired by the late insertion of a resumptive verb form (the item yε), a last resort PF operation that facilitates syntactic-prosodic constituency correspondence. Twi yε-insertion, I claim, is more accurately an instance of lower copy pronunciation, in which a categorially neutral root morpheme that has raised through the head positions of AspP and beyond is spelled out as an impoverished/default verb form. Other vocabulary items compete with yε for insertion (including the lexical verb root from the numeration), but only the insertion of default yε yields a well-formed PF output in the language. (I assume that distinctness conditions on linearization (see Nunes 1999, 2004, Richards 2010) block the realization of a low repeated/nondistinct copy of the original verb root and force the item yε to surface instead.) Furthermore, because an additional (higher) copy of the root morpheme is realized in T0 in these constructions (i.e., as the lexical verb form), Twi verbal resumption can also be analyzed as an instance of multiple copy Spell-Out, an otherwise marked PF outcome.
Because of its PF profile in these respects, Twi verbal resumption contributes to the growing catalogue of conditions known to drive multiple copy realization at the syntax-phonology interface (see Kandybowicz 2008), while deepening our understanding of the nature and inner workings of prosodic mapping. I consider these contributions to be noteworthy given the limited impact of Twi grammar on the theoretical literature thus far and the fact that yε-insertion has remained an otherwise neglected corner of its grammar.
The data presented in this article come exclusively from fieldwork and are presented in the official unified orthography of the language. In keeping with orthographic convention, tone is not marked, though it is discussed in the text when relevant. The following abbreviations are used in the glosses of example sentences: CL.DET – clausal determiner; COMP – complementizer; FOC – focus; HAB – habitual; INAN – inanimate; INT – interrogative particle; NEG – negation; PRF – perfect; PROG – progressive; PST – past; SG – singular.
This article has benefited greatly from the input of numerous people, without many of whom it would not have taken flight: my Twi consultants David Opoku, Isaac Opoku, Peter Owusu-Opoku, and Duke Yeboah; the students of my spring 2013 syntax-phonology interface seminar at the University of Kansas; my colleagues Vicki Carstens, Chris Collins, Claire Halpert, Heidi Harley, Peter Jenks, Arsalan Kahnemuyipour, Ruth Kramer, Michael Marlo, Bridget Samuels, Lisa Selkirk, Harold Torrence, Michael Wagner, and Martina Wiltschko; and the audiences of the 2010 Interfaces Workshop at Syracuse University, ACAL (Annual Conference on African Linguistics) 41, McGill University, the University of Missouri, and the University of Texas at Arlington, where portions of this material were presented. Any errors (orthographic, analytical, or otherwise) are my own.
1 Twi has been argued to be a primarily aspectual language, with only a binary past/nonpast distinction for T heads (Osam 2008, Kusmer 2010). The language expresses most nonpast readings using aspectual morphology, as in (3a–e), but encodes future orientation periphrastically by means of the motion verb bε ‘to come’ in combination with a main verb (see (3f–g)). Future constructions in the language thus appear to be serial verb constructions. Because I am interested here in the consequences of main verb raising to T0 and the creation of phonetically empty syntactic domains, I will not concern myself with the periphrastic future tense in what follows.
2 In Twi, habitual aspect can also be encoded via a phonetically null AspHab head, which has no tonal consequences for the realization of the verb (Paster 2010). Thus, when the verb occupies null-headed Asp0Hab, it surfaces in its underlying form. Because the patterns with respect to yε-insertion do not vary from those in (3) when habitual aspect is zero-marked, I will set aside the case of the zero habitual in this article.
3 A reviewer asks whether the complementizer may be dropped in Twi and if so, whether yε-insertion would be forced. The answer is no. Apart from quotative constructions, Twi complementizers may not be dropped. When V takes a clausal complement, as in (5c–d), yε-insertion is systematically obviated.
5 Even though these analyses agree on the morphological status of yε as a verbal suffix, they make contradictory claims about the phonological implications of the item’s affixal status. For example, Dolphyne (1988) describes an alternation in which yε surfaces as either -yε, -i, or -e, the choice based on harmonization with the preceding stem vowel (see the quotation in the text), while Osam (2003) claims that no such vowel harmony occurs on the yε-form. Ofori 2006 and Paster 2010 also contain descriptions of the vowel alternation in yε. On the basis of my observations, Dolyphne’s characterization appears to be more accurate than Osam’s. The item yε undergoes regular ATR harmony with the preceding verb stem vowel. (See footnote 7.)
6 Note that it is conceptually plausible to analyze verbal vowel lengthening in the past tense (see (1)) as involving the suffixation of a phonetically empty timing slot (Ofori 2006) that is filled by way of linking segmental material from the verb stem (Kobele and Torrence 2006, Paster 2010). In this way, the exponent of T0Past would in fact be a suffix (/-X/), but this would not change the fact that (overt) suffixation in the language is rare.
7 In this article, I will not concern myself with the surface phonological realization of what I am representing as yε, which because of processes involving vowel harmony and epenthesis surfaces sometimes as [e, ë] and other times as [jε], as captured in the description of this passage. Instead, I collapse the two realizations into the single written representation 〈yε〉.
8 In this way, conditional na and past aspectual na pattern alike. Neither can undergo focus movement, which is consistent with their status as functional heads.
9 Despite appearances to the contrary, the past tense morpheme na is not homophonous with the focus marker na. The former bears a High tone, the latter a Low tone.
10 The one exception to this generalization occurs in the negative of perfect constructions, where unlike in simple past constructions that build negative forms by combining perfect a- with Neg0 (see (ia)), the negative is built by combining null T0Past with Neg0 without a mediating perfect aspect marker (Schachter and Fromkin 1968, Saah 1994, Dolphyne 1996, Ofori 2006) (see (ib)). In this case, the verb does in fact surface with a lengthened final segment, despite the presence of Neg0.
Kofi a-n-tɔ bi.
Kofi PRF-NEG-buy some
‘Kofi did not buy something.’
Kofi n-tɔɔ bi.
Kofi NEG-buy.PST some
‘Kofi has not bought something.’
I analyze the blocking of segment-final lengthening on the verb in negative constructions as a consequence of Asp0. When Asp0 is merged in a negative construction (see (ia), (10b)), final lengthening is blocked. When Asp0 is not merged in the narrow syntax (see (ib)), final lengthening may occur. Under an analysis in which final lengthening is contingent on V-T0Past (see below), this would mean that Twi verbs may raise through Neg0, but not through filled Asp0.
11 The EPP impoverishment of na may simply be an interim synchronic property connected to the item’s recent grammaticization from what Osam (2003) has analyzed as a conditional/adverbial particle. The prediction of this speculation would be that over time, as the item evolves into a fully regular tense marker, it would acquire the EPP features typically borne by tense morphemes and at that point in the evolution of the language, subjects would precede na.
12 Unless, of course, the vP-internal subject were to move to a higher vP-external AspP specifier position. This movement would successfully derive the correct linear order of subject before aspect without relying on a low position for Asp0, but at the cost of relying on an unmotivated/ad hoc movement. Furthermore, it is not clear that such a structure/ analysis would successfully account for the impossibility of subject-oriented quantifier float in na-constructions with focused subjects, given that the stranded quantifier would be vP-external following the movement of the subject to Spec,Asp0. For these reasons, I take the low aspect analysis adopted above to be more principled than an alternative analysis that seeks to position Asp0 higher in the clausal hierarchy.
13 Following Smith (1991), Travis (2010) makes a distinction between ‘‘viewpoint aspect’’ and ‘‘situation aspect.’’ Whereas the latter refers to Aktionsart or aspectual verb classes such as Activity, State, and Accomplishment, viewpoint aspect is morphosyntactic (i.e., grammatical) aspect and covers categories such as perfective and imperfective. For Travis (2010), inner aspect is the province of situation aspect, while viewpoint aspect is encoded vP-externally within the inflectional domain of the clause. Thus, the claim that viewpoint aspectual categories like perfect, progressive, and habitual are encoded vP-internally in Twi may appear inconsistent with Travis’s proposal. On the contrary, although inner aspect is primarily related to situation aspect in her system, Travis concedes that in some languages the vP-internal aspect projection may host material that is more viewpoint-oriented than situational. As examples of such cases, Travis cites Navajo aspectual morphology, progressive reduplicants in Tagalog, and completive suffixes in Tagalog and Malagasy. (See Travis 2010:chaps. 3 and 7 for detailed discussion of these cases.) Twi thus appears to be another such language that encodes (at least some) viewpoint aspect vP-internally.
14 In this way, the presence of yε tracks the phonological emptiness of AspP. Other languages have different mechanisms for tracking the content of this constituent. One example is Zulu, which utilizes a morphological alternation on the verb to encode whether a verb phrase has content at PF or not, apart from the verb. In Zulu, the so-called ‘‘disjoint’’ verb form tracks empty verb phrases. It appears when both subjects and objects have vacated the verb phrase, leaving behind an empty constituent apart from the verbal head. The ‘‘conjoint’’ verb form, on the other hand, is used whenever one of these operations fails to occur, that is, when the verb phrase has PF content beyond the verb. Of course, this description is somewhat oversimplified. See Buell 2006, 2008 for more on the Zulu conjoint/disjoint alternation and how it can be pressed into service to probe whether a given base-generated vP constituent is internal or external to vP at PF.
15 Although upper middle field modifiers such as subject-oriented adverbs like ‘foolishly’ or ‘confidently’ would be predicted to force yε-insertion on this analysis, they do not have the same postverbal distribution as the adverbials considered above. Rather, they appear sentence-initially and therefore do not play a decisive role in distinguishing the present analysis from Osam’s linear approach.
A reviewer asks whether the low adverbs ntεm ‘quickly’ and komm ‘quietly’ allow both subject-oriented and manner readings, as in English, and if so, whether only the manner readings of such adverbs obviate yε-insertion, as the present analysis predicts. While the predictions of the current proposal are clear and unambiguous, I unfortunately lack the data to answer this question, and I leave it for future research.