4.1 The clause-typing particles

Tuparí distinguishes different clause types using a set of 7 second position particles, listed in (8). Of these, =nẽ, =nãpe and =’aet were classified as members of a broader set of particles by Caspar and Rodrigues (1957, §3.5), though their syntactic distribution was not explored in detail there.

We must distinguish between two homophonous clause-typing particles: mãkẽrõ ‘dont.know’ and mãkẽrõ ‘confirmative’. They have distinct prosody (mãkẽrõ ‘confirmative’ is accompanied by a sharply rising intonation) and trigger different effects on the deictic orientation of the evidential suffix -pnẽ/psira (Singerman 2019, 411–428).

(9) shows the semantic effects that these clause-typing particles can have.

An important fact about the clause-typing particles is that they only ever combine with fully-formed utterances. A clause that lacks an overt clause-typing particle is not ill-formed; it just instantiates a superficially unmarked sentence type. Hence the counterpart to (9a) through (9c) without an overt clause-typing particle is just a run-of-the-mill declarative, without any special emotive or emphatic content: Kafet te’a ‘There is coffee’. That the clause-typing particles are added to otherwise fully-formed utterances supports an analysis on which these particles sit in a high position in the clausal spine, one above TP, EvidP, and the two auxiliary projections shown in Fig. 2. If this projection may host one and only one constituent in its specifier, as shown in Fig. 3, then we also explain why the clause-typing particles consistently surface in second position.

View Full Size

The highest level of the Tuparí clause contains a head-initial CP

Citation: Acta Linguistica Academica 67, 4; 10.1556/2062.2020.00025

• Download Figure
• Download figure as PowerPoint slide

4.2 What kind of constituents can precede the second position clause-typing particles?

It is a well-established fact that morphemes may end up in second position by a variety of mechanisms. Second position effects – including Verb Second – bear the hallmarks of head movement in many languages, including German (den Besten 1983), Kashmiri (Bhatt 1999; Manetta 2011, 2020) and, within the Tupían family, Karitiana (Storto 1999, 2003, 2020). In other cases, however, phonological and/or prosodic mechanisms are needed to account for second position effects. This is the motivation behind Halpern's (1995) operation of Prosodic Inversion and behind Anderson's (2005) constraint-driven model of clitic placement. This subsection presents evidence that the placement of the clause-typing particles in second position is a syntactic phenomenon: the constituents that occur prior to the second position particle cluster in this language are not amenable to phonological or prosodic characterizations but must instead be defined in terms of syntactic integrity. To keep things easier on the reader we prioritize examples that contain =nẽ ‘yes/no’, one of the most common clause-typing particles. However, all of the clause-typing particles (listed in 8) exhibit identical behavior with regard to syntactic constituency.

The clause-initial constituent can be an NP subject, shown in (10) by the simplex uokat ‘water’ and by the slightly more complex Kopere ema’en ‘language of the Djeoromitxí people’. Observe that all non-pronominal subjects are required to bear the nuclear case -et/en/t/n.

Fully finite embedded clauses nominalized by hè can also serve as NP subjects (see Singerman 2020 for more on these clausal nominalizations, which are a diachronic innovation unique to Tuparí). In (11), the clausal nominalization immediately precedes the matrix clause-typing particle =nẽ ‘yes/no’.

The embedded clause in (11) is fully finite: tarupat teotsirat hèt ‘the white people who went (non-witnessed)’ contains the plural evidential suffix -psira and the near past suffix -t. Note that this nominalization is interpreted as an internally headed relative, one of the major functions of this language's finite embedded clauses (For more on IHRCs, see Cole 1987; Williamson 1987; Basilico 1996; Hiraiwa 2017; Ohara 2018; Hanink 2020).

The clause-initial constituent can also be a VP. Intransitive lexical verbs bear a pronominal proclitic that agrees with the subject in number and person (e- ‘2sg’ in 12a). Transitive lexical verbs are immediately preceded by the direct object.⁵ This is true for pronominal objects (third person y- in 12b) and non-pronominal objects (ositot ‘my foot’ in 12c) alike.

The direct object in a clause-initial verb phrase can also be a clausal nominalization headed by hè.

Like tarupat teotsirat hèt ‘the white people who went (non-witnessed)’ in (11), the bolded constituents in (13) are interpreted as internally headed relatives: Elizanin nemnan hèt ‘the thing that Elizane made (non-witnessed)’, where the internal head is a null pronominal object; ’àpère i’anam kiret hèt ‘the people who live alongside the road’, where the internal head is the subject NP kiret ‘the people’. In both of these examples, the transitive matrix verb together with the object serve as the clause-initial constituent. In this sense these data are fully parallel to (12b) and (12c), despite the intuitive difference in weight: whereas the object in (12b) is a mere pronominal proclitic and that in (12c), a possessed noun, in (13a) and (13b) the direct objects are fully finite clausal nominalizations.

Other kinds of constituents, such as adverbials, can also serve as the clause-initial constituent. Here, too, we observe the same range of possibilities already seen in (10) through (13): the clause-initial adverbial may be very small (just monomorphemic) or very large (say, an entire finite embedded clause).

In (14a) the clause-initial adverbial – eret ‘tomorrow’ – is monomorphemic. In (14b) the clause-initial adverbial is structurally more complex, containing a lexical verb – apsitkòmkat ‘grow sad’ – as well as the subordinating morpheme (’)a ‘if, when’. (Both the lexical verb and the subordinator are preceded by the second person singular prefix e-.) And in (14c) the clause-initial constituent is a clausal nominalization, headed by hè, that itself contains a direct quotation: apait teynã ke y’e Adãõn ‘Adam said that auntie came (non-witnessed)’.⁶

We must emphasize that the second position facts shown above are consistent across speakers, independent of gender, knowledge of Portuguese, or other sociolinguistic factors. Speakers never interrupt clause-initial constituents with second position particles; they always respect those constituents' internal integrity. In short, what matters for the placement of the clause-typing particles is syntactic constituency. We therefore maintain that the tree given in Fig. 3 – where a single constitutent may occupy Spec,CP, thus yielding the placement of the clause-typing particles in second position – is the right representation for the highest layer of the Tuparí clause.

4.3 The clause-typing particles as C heads

The projection headed by the clause-typing particles is given as C in Fig. 3. Motivation for this label comes from the fact that the clause-typing particles are sensitive to whether or not the clause-initial constituent is [$+wh$], just like complementizers in better studied languages (such as English if, whether and that). Only =nãpe ‘emphatic’ and =mãkẽrõ ‘dont.know’ may occur with [$+wh$] clause-initial constituents.

Analyzing the clause-typing particles as heading C, as we have done here, is further supported by the fact that all $wh$-words in Tuparí must occur clause-initially; $wh$-in-situ is categorically unattested. (See Brandon and Seki 1981, 1984 for early discussion of $wh$-movement in Tupían languages.) The $wh$-word will immediately precede an overt clause-typing particle in the linear string.⁷

When a $wh$-word is the complement of a postposition or the possessor of another noun, the entire phrase will pied-pipe.

The sensitivity of the various clause-typing particles to the [$\pm wh$] status of the clause-initial constituent can be interpreted as a reflex of the $wh$-criterion (Rizzi 1996). On this analysis the particles are simply complementizers, each specified as compatible or incompatible with a [$+wh$] XP in their specifier. More specifically, =mãkẽrõ ‘dont.know’ requires a [$+wh$] constituent to sit in Spec,CP; =nãpe ‘emphatic’ accepts a constituent that is [$+wh$], though it will also accept a [$-wh$] constituent; and the remaining particles are all incompatible with [$+wh$] clause-initial constituents.

4.4 Summary

The data discussed in this section have demonstrated that it is syntactic constituency that determines the placement of the clause-typing particles in second position. Analytically we have implemented this finding by positing a head-initial CP at the top of the Tuparí clause, as in Fig. 3; this projection will host a single XP in its specifier, giving rise to the clause-typing particles' placement in second position. So while the lower levels of the Tuparí clause show clear head-finality, at the top we find unambiguously head-initial phrase structure. In the next section we turn to the Tense Phrase, which is located between these two zones of headedness and which shares properties with both of them.

5.1 An overview of tense marking in Tuparí

The full set of tense marking in Tuparí is heterogenous and consists of three mutually exclusive sets of morphemes: (a) predicate-final suffixes, (b) enclitic particles located in second position, and (c) auxiliaries that follow the lexical VP. Despite this heterogeneity, all three sets of morphemes behave comparably in discourse and serve the same general semantic function – namely, to anchor the event time with respect to the Utterance Time (UT). For reasons of space this paper focuses on the predicate-final suffixes and the second position particles. For more information on the post-VP auxiliaries that mark tense and aspect, the reader is referred to Singerman (2018b, ch. four).

The paradigm in (18) gives an illustration of the predicate-final suffixes and second position particles, which paradigmatically oppose one another. In each of the four utterances, the relevant tense morphology is highlighted, as is the second position clause-typing particle =mãkẽrõ ‘dont.know’.

We have highlighted the clause-typing particles in these five examples to emphasize that whereas the tense particles will occur in the second position particle cluster – immediately following the clause-typing particle – the predicate-final tense suffixes may be realized far from second position. As shown in (18b), the predicate-final suffix will attach to the auxiliary when one is present; in the absence of an auxiliary, as in (18a), the suffix attaches to the lexical verb itself.

The second position tense particles and predicate-final tense suffixes must be analyzed as constituting a single grammatical class. They all carry out the same semantic function: they anchor the event time with respect to the utterance time. As indicated by the translations in (18), the near past suffix -t/n, the distant past particle =õpot and the ancient past particle =kut divide the before-now timeline into carefully delineated intervals: -t/n is used for events that took place at least two days before UT through several months into the past; =õpot, for events from approximately two years before UT to the speaker’s early childhood; and =kut, for events that took place prior to or at the speaker’s birth. The predicate-final durative suffix -pbi’a/mbi’a, meanwhile, conflates tense, aspect and evidentiality: it is used for non-future habitual actions that the speaker has witnessed firsthand. Speakers are very strict about the meaning of each of these morphemes. Misusing them gives rise to inaccurate temporal interpretations (and aspectual/evidential ones, in the case of durative -pbi’a/mbi’a); these erroneous uses are in turn subject to correction by native speakers. And as demonstrated by the textual data in Singerman (2018b, 298–303), all of this tense morphology behaves comparably in running discourse.

A further reason to analyze the second position tense particles and the predicate-final tense suffixes as instantiating the same overarching grammatical category is that members of both sets of morphemes can combine with the aux_go auxiliary series to introduce intermediate temporal gradations. The near past suffix -t in (18a) marks the hammock-hanging event as having taken place between two days and several months before UT; but the combination of -t/n ‘near.past’ with the aux_go in (18b) yields a more removed temporal interpretation. In a comparable fashion, the second position tense particle =õpot in (18c) marks the hammock-hanging event as having taken some two years prior to UT; and the addition of the aux_go in (18d) pushes the temporal interpretation back even farther.

The difference between the pair in (18a) and (18b), or between the pair in (18c) and (18d), is not akin to that between English went and had gone. English-style pairs of past and past perfect indicate the relationship that two different past tense events have to one another (see Klein 1992). In Tuparí, however, the combination of a predicate-final suffix or second position particle with a member of the aux_go series is part of the language's nuanced tense system: these combinations are used to introduce intermediate temporal gradations, all of which are anchored to the UT. Consider (19). These five lines are excerpted from a text that relates a series of events which took place more than just a few months ago (and therefore too far back for the narrrator to use the near past suffix -t/n) but less than two years ago (and therefore too recent for him to use the distant past second position particle =õpot). Each finite clause in this excerpt contains both near past -t and a member of the aux_go series: paucal oro’e in (19a), singular tero’e in (19b) through (19e).

Just like the predicate-final suffix -t/n ‘near.past’, the second position particle =õpot ‘distant.past’ will combine with the aux_go series to produce more nuanced temporal gradations. Example (20a) is how an elderly woman described being breastfed by her grandmother, her mother having passed away soon after her birth. The speaker of this example is around eighty years old, and consultants confirm that both the distant past particle =õpot and the auxiliary tet’e are obligatory here. Using =õpot without tet’e, as in (20b), would work only if someone much younger – say, a young child – were to relate an event that had taken place just a few years before UT.⁸

In sum, there are multiple reasons to analyze the predicate-final tense suffixes and second position tense particles as belonging to a single grammatical class. First, with the exception of the particle =ko/ke ‘polite.future’, all of these morphemes divide up the prior-to-UT timeline in comparable fashion. Second, they all may combine with the same auxiliary series to produce intermediate temporal gradations. Third, these morphemes behave identically in actual discourse; running texts see the right tense morpheme repeated in every clause. This was shown for the near past suffix -t/n in (19): each finite clause in that textual excerpt contains both the near pst suffix and a member of the aux_go series. Singerman (2018b, 298–303) provides comparable textual evidence for distant past =õpot, ancient past =kut, and durative -pbi’a/mbi’a. As these morphemes all share comparable semantics, enjoy similar distribution in discourse, and are mutually exclusive, they must be analyzed as realizing the same overarching grammatical category.

5.2 What is the underlying headedness of the Tuparí Tense Phrase?

Having argued that the predicate-final suffixes -t/n ‘near.past’ and -pbi’a/mbi’a ‘durative’ and the second position particles =ko/ke ‘polite.future’, =õpot ‘distant.past’ and =kut ‘ancient.past’ all instantiate T, we must ask about the headedness of the Tense Phrase. To consider the different analytical possibilities, consider the four-member past tense paradigm given in (18). In (18b), the near past suffix -t attaches to the auxiliary root tero’e ‘aux_go.sg’, which itself follows the VP ewaet ãpea. This VP is head-final, too, with the direct object ewaet ‘your hammock’ preceding rather than following the transitive verb ãpea ‘hang up’. In this sense T in (18b) is head-final, just like the other projections below it: the VP and the AuxP. In (18c) and (18d), on the other hand, the distant past particle =õpot sits in second position, immediately after the clause-typing particle =mãkẽrõ ‘dont.know’. There is no evidence in (18c) or (18d) of T behaving comparably to the head-final VP or head-final AuxP; rather, it behaves like the C projection, which we argued in Section 4 to be head-initial.

One could imagine three possible lines of analysis. T, like C, could be underlyingly head-initial, such that the surface position of the predicate-final tense suffixes is the result of some dislocation operation. On this analysis, it is the second position particles that reflect the underlying headedness of the TP. Alternatively, T could be head-final, like the EvidP and other lower projections, such that the appearance of the tense particles in second position results from dislocation. On this analysis, the predicate-final suffixes, not the second position particles, are what reflect the TP's underlying headedness. Finally, it is possible that neither the second position tense enclitics nor the predicate-final tense suffixes reflect the underlying headedness of the Tense Phrase. We will argue for this third analysis: both sets of morphemes undergo some form of dislocation to reach their surface position, such that neither the predicate-final suffixes nor the second position particles provide an unfiltered view of the underlying headedness of the TP.

5.2.1 The second position particles undergo dislocation to their surface position

Recall that the second position tense particles always occur immediately after a clause-typing particle (if one is present). Inverting the linear order of the two particles is completely unattested. Several examples containing both an overt clause-typing particle and an overt tense particle were already given in Sections 4.2 and 5.1 (21) provides three more.

(21)	The clause-typing particle immediately precedes the tense particle in second position
	a.	Erero	ta	ke	’en	ipekap.
		erero	=ta’a	=ke	=’en	i-pek-ap
		early.tomorrow	=assertive.♀	=polite.future	=2sg	3-buy-adv.foc
		‘Please buy it early tomorrow morning.’
		conversation: 2018-08-31

	b.	En	nãkop	õpot	’en	Marli wãkiret	kat.
		en	=nãkop	=õpot	=’en	Marli wãkit-et	ko-a-t
		2sg	=maybe	=distant.past	=2sg	Marli pet-nuc	eat-th-nuc
		‘Perhaps it was you who long ago ate Marli’s pet.’
		conversation: 2018-08-07

	c.	Pare	mãkẽrõ	kut	yan	osinã	tet’e.
		pare	=mãkẽrõ	=kut	yã-n	o-sinẽ-a	tet’e
		where	=dont.know	=ancient.past	mother-nuc	1sg-give.birth-th	auxgo.sg
		‘I do not know where my ormmother gave birth to me.’
		conversation: 2017-08-30

That the tense particle immediately follows the clause-typing particle in the second position cluster can be accounted for by T-to-C Head Movement (for discussion of Head Movement see Travis 1984; Matushansky 2006; Harley 2013; Zeller 2013; McCloskey 2017, among many others). T and C will thus form a single complex head which follows the clause-initial constituent. Positing T-to-C Head Movement in this circumstance successfully captures the fact that in utterances that contain both a clause-typing particle and a second position tense enclitic, they are always adjacent; to our knowledge, no other material is ever capable of surfacing between them.

However, if the tense particles arrive in second position via T-to-C Head Movement, then an underlyingly head-final TP and an underlyingly head-initial TP will both serve equally well: the original headedness of the TP will be hidden by the application of Head Movement. Put slightly differently: if T-to-C Head Movement always applies to ensure that the tense particles surface in second position, then these particles will tell us nothing about the TP's underlying headedness. This is illustrated by Figs 6 and 7.⁹

View Full Size

Tense undergoes Head Movement to C from head-initial TP

Citation: Acta Linguistica Academica 67, 4; 10.1556/2062.2020.00025

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Tense undergoes Head Movement to C from head-final TP

Citation: Acta Linguistica Academica 67, 4; 10.1556/2062.2020.00025

• Download Figure
• Download figure as PowerPoint slide

A comparable situation obtains in the Germanic languages, which are standardly analyzed as displaying Head Movement from T to C; it is this operation that accounts for the Verb Second (V2) pattern. But the headedness of the T and V projections varies among the Germanic languages: for example, these projections are standardly analyzed as head-final in German but as head-initial in English (See den Besten 1983; Haider and Prinzhorn 1986; Vikner 1995 as well as various contributions to Wolfe and Woods 2020.). Thanks to the fact that the C projection is consistently head-initial across Germanic, T-to-C Head Movement will yield V2 surface order (or, in the absence of an initial constituent, verb-initial order) regardless of the underlyingly headedness of the TP or VP. In much the same way, the obligatory Head Movement of the second position particles =ko/ke ‘polite.future’, =õpot ‘distant.past’ and =kut ‘ancient.past’ from T to C means that these particles are uninformative as to the underlying headedness of the Tuparí TP.

A reviewer points out that recent scholarship has problematized the traditional conception of Head Movement: what was previously viewed as a unitary process may in fact consist of several distinct operations, each with its own set of properties. In particular, Harizanov and Gribanova (2019) propose to bifurcate classic Head Movement into (a) Internal Merge, which applies to full words and belongs to the core syntactic component, and (b) Head Raising, which is a postsyntactic operation involved in word-building ($=$morphological) processes. This bifurcation allows Harizanov and Gribanova to explain those cases of Head Movement that have been argued to have interpretive consequences (see for instance Keine and Bhatt 2016 on the German construction known as the Long Passive). On Harizanov and Gribanova's typology, such cases are the product of Internal Merge; that is, they must apply within the narrow syntactic component. Cases which do not affect interpretation, on the other hand, may result either from Internal Merge or from the postsyntactic operation of Head Raising.

The reviewer asks where the Tuparí facts fit into the refined typology of Harizanov and Gribanova. Do Figs 6 and 7 illustrate an application of Internal Merge, which by definition must take place within the narrow syntax, or an application of postsyntactic Head Raising? As a preliminary answer to this question, observe that the semantic contribution of the different tense morphemes is unconnected to whether they are realized as a predicate-final suffix (near past -t/n, in examples 18a and 18b) or as a second position particle (distant past =õpot, in 18c and 18d). There is zero correlation in Tuparí between where tense morphology is ultimately realized in the clause and what kind of semantic contribution that morphology provides. This lack of semantic consequence is compatible in principle with either the Internal Merge or Head Raising operations, since on Harizanov and Gribanova's theory Internal Merge may yield semantic effects but need not do so. More decisive evidence comes from the fact that the second position particles =ko/ke ‘polite.future’, =õpot ‘distant.past’ and =kut ‘ancient.past’ do not constitute fully-formed words in their own right; rather, they are bound, destressed morphemes that must encliticize onto preceding material. (That material will be a clause-typing particle, if one is present, though the clause-typing particle in turn must encliticize onto the clause-initial XP. If there is no overt clause-typing particle, then the tense particle will lean onto the clause-initial XP directly.) For Harizanov and Gribanova, Internal Merge applies to fully-formed words while Head Raising may be driven by purely morphological considerations. The operation that best accounts for the realization of certain tense morphology as second position enclitic particles in Tuparí would thus be Head Raising (with right adjunction to ensure that the tense particle follows rather than precedes the clause-typing particle).

The next subsection discusses what the suffixes -t/n ‘near.past’ and -pbi’a/mbi’a ‘durative’ can reveal about the underlying headedness of the Tense Phrase. These suffixes arrive in their surface position, at the right edge of the predicate, via a Lowering operation. As a result, they are also uninformative with regards to whether the Tuparí TP should be analyzed as underlyingly head-initial or underlyingly head-final.

5.2.2 The predicate-final suffixes also undergo dislocation to their surface position

The surface position of the predicate-final suffixes does not tell us anything more about the underlying headedness of the TP than the second position particles do. This is because of the nature of predicate fronting in Tuparí. When a lexical verb cooccurs with one or more auxiliaries, only the VP headed by the lexical verb may occur to the left of a second position clause-typing particle. The auxiliary or auxiliaries will surface to the right of the clause-typing particle. As a result of the fact that the tense suffix -t/n ‘near.past’ always attaches to the highest verbal head, in examples like (22a) the clause-typing particle will linearly separate the lexical VP from the auxiliary that bears the tense suffix.

(22)	Placement of clause-typing particles in clauses with auxiliaries
	a.	Itoa	nẽ	etero’at	en?
		i-top-a	=nẽ	e-tero’e-a-t	=’en
		3-see-th	=yes/no	2sg-auxgo.sg-th-near.past	=2sg
		‘Did you see it?’
		elicitation: 2020-02-02

	b.	*Itoa	etero’at	nẽ	’en?
		i-top-a	e-tero’e-a-t	=nẽ	=’en
		3-see-th	2sg-auxgo.sg-th-near.past	=yes/no	=2sg
		(intended to mean the same as 22a)
		elicitation: 2020-02-02

The crucial observation is that in (22a) only the lexical VP itoa occurs prior to the clause-typing particle =nẽ ‘yes/no’. The auxiliary, meanwhile, surfaces to the right of the clause-typing particle; and it is to the auxiliary that the near past suffix attaches. As (22b) demonstrates, placing both the lexical VP and the tense-marked auxiliary prior to =nẽ ‘yes/no’ results in ungrammaticality. The contrast between (22a) and (22b) shows that a full TP cannot front to the clause-initial position, which we identified in Section 4 as Spec,CP. But now consider the examples in (23), where – due to the lack of an auxiliary – the tense suffix surfaces on the lexical verb itself. Here, the lexical VP occurs to the left of the second position clause-typing particle =nẽ ‘yes/no’.

(23)	Fronted VPs that themselves carry tense morphology
	a.	E’ẽkapbi’a	nẽ	’en	mõket?
		e-’ẽk-a-pbi’a	=nẽ	=’en	mõket
		2sg-dance-th-durative	=yes/no	=2sg	long.ago
		‘Did you used to dance?’
		elicitation: 2018-08-23
	b.	Epuop’orat	nẽ	'en	haytokia?
		e-puop’ot-a-t	=nẽ	=’en	haytokia
		2 sg-learn-th-near.past	=yes/no	=2sg	a.lot
		Did you learn a lot?
		elicitation: 2018-09-01

Example (22a) would appear to contradict the two utterances given in (23): according to the former the fronted constituent is smaller than TP, since -t/n ‘near.past’ surfaces on the auxiliary, which remains to the right of the clause-typing particle in the linear string; but in the latter, the fronted VP itself bears the tense suffix, thereby suggesting that a constituent larger than just the VP itself has fronted to Spec,CP.

To resolve this contradition we propose that utterances such as (28a) and (23b) feature Lowering (Harley and Noyer 1999; Embick and Noyer 2001; Harizanov and Gribanova 2019) from T to the next highest syntactic head in the clausal spine. This head may be an evidential suffix, the structurally highest auxiliary, or the the lexical verb itself. In this last case a lexical verb that bears tense morphology can front to the clause-initial position ($=$Spec,CP), as in (28a) and (23b).

But just as Figs 6 and 7 demonstrated for T-to-C Head Raising, the underlying headedness of the TP is irrelevant once Lowering applies. A head-final TP (as in Fig. 8) or a head-initial one (Fig. 9) will serve equally well, since both satisfy the head-to-head locality condition that is hypothesized to characterize postsyntactic head-displacement operations. In sum, the predicate-final tense suffixes do not help to disambiguate the basic headedness of the TP any more than the second position tense particles do. Rather, both sets of tense morphemes are dislocated to their surface position.

View Full Size

Tense suffix undergoes postsyntactic Lowering to V from head-final T

Citation: Acta Linguistica Academica 67, 4; 10.1556/2062.2020.00025

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Tense suffix undergoes postsyntactic Lowering to V from head-initial T

Citation: Acta Linguistica Academica 67, 4; 10.1556/2062.2020.00025

• Download Figure
• Download figure as PowerPoint slide

Here we find an interesting parallel to the facts surrounding the second position particles =ko/ke ‘polite.future’, =õpot ‘distant.past’ and =kut ‘ancient.past’, which were discussed in the prior subsection. That subsection argued that if we assume the bifurcation of Head Movement proposed by Harizanov and Gribanova (2019), then the second position particles must reach their surface position via an application of the postsyntactic operation of Head Raising rather than the narrowly syntactic operation of Internal Merge. As evidence in favor of a postsyntactic rather than narrowly syntactic account for the second position particles' behavior, we observed that the particles themselves do not qualify as independent words but are instead phonologically dependent enclitics. The same logic applies to the predicate-final suffixes, too: these morphemes are not independent words but rather bound affixes that must attach to a verbal stem.¹⁰ We thus arrive at a picture of Tuparí tense morphology where one of two different postsyntactic operations will apply, depending on the specific piece of morphology that is present. In the case of the second position particles, that operation will be Head Raising from T to C. In the case of the predicate-final suffixes, the operation will be Head Lowering from T to the head of the next-highest projection in the clause (which may be the Evidential Phrase, an Auxiliary Phrase, or the VP itself).¹¹

5.3 Summary

This section has examined the headedness of the Tuparí Tense Phrase, which sits above a zone of head-finality and beneath a zone of head-initiality. Recent research into disharmonic syntax has argued that head-initiality works ‘top down’ and head-finality ‘bottom up’ within Extended Projections (in the sense of Grimshaw 2000, 2005). If only one head-initial phrase occurs within an Extended Projection, it will be the highest one; if only one head-final phrase occurs within an Extended Projection, it will be the lowest. This observation has grown out of a range of scholarship on Final-over-Final Condition/Constraint, first proposed in Holmberg (2000) and later relativized to Extended Projections by Biberauer et al. (2014) (see also Sheehan et al. 2017). Main clauses in Tuparí conform to FOFC in that clearly head-final phrase structure is found at the bottom of the clause whereas unambiguous head-initiality occurs only at the top.¹² In between these two domains we encounter the murky, indeterminate TP, indicated in Fig 10 with dotted lines.

View Full Size

The Tuparí Tense Phrase is of indeterminate underlying headedness

Citation: Acta Linguistica Academica 67, 4; 10.1556/2062.2020.00025

• Download Figure
• Download figure as PowerPoint slide

This kind of clausal organization is likely widespread within the Tupían family, as various Tupían languages have head-final VPs but exhibit V2/second position effects alongside obligatory $wh$-movement to the left periphery. See, among others, Brandon and Seki (1981, 1984) on Kamaiurá, Moore (1984) on Gavião, Gabas Jr. (1999) on Karo, and Storto (1999, 2014) on Karitiana.

6.1 The distribution of the weak nominative enclitics

Table 1 shows the seven weak nominative enclitics (see Singerman 2018b, 29–42 for discussion of their morphophonology). These enclitics are subject to the following requirement: they must occur immediately after the tense morpheme, whether this is a second position particle or predicate-final suffix. The crucial factor that determines the availability and placement of the weak nominative enclitics is the kind of tense marking present. (It is for this reason that we can only describe these enclitics now, following Section 5's discussion of the headedness of the TP.)

In utterances where T is realized as a second position particle, the weak nominative enclitic must immediately follow it. The minimal pair in (25), offered back-to-back in conversation by an elderly speaker, illustrates.

In (25a) the clause-initial constituent is the verb wepsika; the postpositional phrase omemsĩremsin yare occurs to the right of the second position cluster. In (25b) that postpositional phrase now serves as the clause-initial constituent, while the VP wepsikap (bearing special adverbial focus morphology) follows the second position cluster. Yet in both examples =’on ‘1sg’ occurs to the immediate right of =ko ‘polite.future’. The syntactic category of the clause-initial constituent and the linear location of the lexical verb are irrelevant for the placement of the weak nominative enclitic; the only important factor is the presence of the particle =ko ‘polite.future’. As this particle occurs in second position, first person singular =’on must also do so.

Separating the weak nominative enclitic from the tense particle leads to ungrammaticality.

This pair of ungrammatical utterances demonstrates that no material may intervene between a second position tense particle like =ko ‘polite.future’ and a weak nominative enclitic.

In the case of the predicate-final tense suffixes -t/n ‘near.past’ and -pbi'a/mbi'a ‘durative’, the weak nominative enclitic will surface immediately after the suffix. The clause-typing particles and weak nominative enclitics are highlighted in (27) so as to emphasize that while the clause-typing particles invariably occur in second position (just like the particles =ko/ke ‘polite.future’, =õpot ‘distant.past’ and =kut ‘ancient.past’), the weak nominative enclitics do not.

In both of these examples an overt clause-typing particle sits in second position, thereby demarcating the right boundary of the clause-initial constituent (pare ‘where’, Guarani’eat yam ‘to the Guarani’). But the weak nominative enclitics do not appear in second position; they instead follow the tense suffix (-t/n ‘near.past’, -pbi’a/mbi’a ‘durative’). So there exists no direct connection between the weak nominative enclitics and second position. Rather, the position of the weak nominative enclitics depends on that of tense – a grammatical category that is itself split in realization between predicate-final suffixes and second position particles. Whenever a weak nominative enclitic occurs in second position, it is because tense morphology is there, too. (In Section 7 we will use this generalization to argue that the position of weak nominative enclitics in superficially tenseless clauses provides evidence for the presence of a null T head.)

Example (26), above, showed that the weak nominative enclitic cannot be linearly separated from a second position tense particle. The pair in (28) shows that the weak nominative enclitic cannot be linearly separated from a predicate-final tense suffix, either.

There is only one morphosyntactic configuration where the weak nominative enclitic can be linearly separated from tense morphology: when both (a) a tense-marked VP occurs in clause-initial position and (b) a clause-typing particle is present.

These facts are discussed at greater length in the next subsection.

The close relationship between the weak nominative enclitics and tense morphology is illustrated by a small set of portmanteaux. The shape of the polite future particle varies according to the person and number of the subject: =ko is used with the first person singular and first person plural exclusive and =ke is used with the second and third persons, regardless of number. Yet special portmanteaux appear when the subject is first person dual inclusive or first person plural inclusive.

Given that the first person inclusive enclitics are otherwise ’okit and ’okitwat, we would expect *ko ’okit in (30a) and *ko ’okitwat in (30b); but the only possible forms are kit and kitwat, respectively. This conflation of tense together with person and number features of the subject suggests analyzing the weak nominative enclitics as a kind of agreement on T rather than as true pronominal arguments, contra Alves (2004).

The weak nominative enclitics highlighted in (25) through (30) are not optional: they cannot be removed without rendering the utterances ungrammatical. But whereas these enclitics are obligatory with the predicate-final tense suffixes and with the second position tense particles, they are prohibited from cooccuring with the auxiliaries that provide present, future, and same-day past interpretations (see Singerman 2018b, ch. four for paradigms and examples). Hence utterances like the following four never include weak nominative enclitics.

The absence of the weak nominative enclitics in these utterances is due to the fact that the right sort of tense morphology is missing. That is, these utterances lack a predicate-final tense suffix or second position tense particle, on which the weak nominative enclitics are parasitic.

To summarize, the weak nominative enclitics (a) may only occur with a specific subset of tense morphology (the predicate-final suffixes and second position particles, but not the post-VP auxiliaries); (b) must follow the tense morphology in the linear string; and (c) in at least one case are realized via portmanteaux with tense. All of these factors suggest that these morphemes are not actual pronouns, as claimed in prior descriptive work on Tuparí, but instead instantiate agreement between the subject and the T head. If =’on ‘1sg’, =’en ‘2sg’, and the other weak nominative enclitics were true subjects it would be difficult to make sense of their restriction to a particular subset of T heads. If, however, these enclitics are some kind of agreement morphology, their circumscribed distribution is less worrisome. Recent work on the agreement/argument distinction exploits this same logic: Nevins (2011) and Kramer (2014) argue that insensitivity to T/INFL is diagnostic of argument clitics rather than agreement morphology. In that Tuparí weak nominative enclitics are parasitic on certain kinds of tense marking, we have good reason to analyze them as agreement rather than true arguments.

6.2 Concerning the surface position of the weak nominative enclitics

Section 5 argued that the second position tense particles and the predicate-final tense suffixes arrive in their surface positions via postsyntactic dislocation operations: Head Raising and Head Lowering, respectively. If the morphemes referred to here as weak nominative enclitics are the realization of agreement between the T head and the subject, we must ask what mechanism is – or mechanisms are – responsible for ensuring that the weak nominative enclitics always follow the tense morphology in the linear string.

I assume here that the weak nominative enclitics reflect the application of the standard Agree mechanism utilized in much Minimalist research (Chomsky 1995). The T head will have unvalued features for person and number; these features are valued via the application of Agree between T and the subject, which is base-generated in Spec,VP. Once the Tense head's features for person and number are valued by Agree, they can be realized as the appropriate member from the set of enclitics in Table 1. In the case of the polite future portmanteaux (examples 30a and 30b), specific spell-out rules will fuse the tense particle and the weak nominative enclitic into a single form, one incapable of further morphological segmentation. In all other cases, the weak nominative enclitic will be spelled out as separate from, but directly attached to, the tense morpheme. If the weak nominative enclitic and T together form a single complex head, they will be capable of undergoing the dislocation operations of Head Raising (which places the particles =ko/ke ‘polite.future’, =õpot ‘distant.past’ and =kut ‘ancient.past’ in second position) and Head Lowering (which attaches the suffixes -t/n ‘near.past’ and -pbi’a/mbi’a ‘durative’ to the right edge of the predicate). Since the weak nominative enclitic will move with the tense morphology as it undergoes Head Raising / Head Lowering, it will always remain adjacent to tense.

Although I have referred to these morphemes as enclitics, they could plausibly be treated as suffixes – suffixes that are subcategorized to attach directly on top of tense morphology. (As will be discussed further in Section 6.4, below, comparative evidence indicates that the Tuparí weak nominative enclitics are partway through a grammaticalization process from independent pronoun to enclitic pronoun to agreement affix.) There is only one phenomenon that mitigates against treating the weak nominative enclitics as synchronic suffixes: they may be linearly separated from a tense suffix by an intervening clause-typing particle. This was shown in (29) for both the near past and durative suffixes. In no other circumstance may the weak nominative enclitic be separated linearly from a tense suffix; what is more, I know of no circumstance where a weak nominative enclitic may be linearly separated from a second position tense particle. Were it not for examples such as (29), then, there would be no evidence against an analysis of the weak nominative enclitics as agreement suffixes glued directly onto the T head. I therefore propose that utterances such as (29a) and (29b) be treated as the product of a special adjustment rule, one that applies late in the derivation and takes as its input the linear order of the morphemes.¹³

The input for this rule makes reference to a hypothesized intermediate step where the clause-typing particle follows rather than precedes the weak nominative enclitic. Such an intermediate step would be derived straightforwardly given the other operations already argued for. In particular, Agree will lead to the valuation of the person and number features on the T head, giving rise to the weak nominative enclitic as an enclitic or suffix that immediately follows the tense morphology. Subsequently, the operation of Head Lowering will bring the complex head consisting of T and the weak nominative enclitic down onto the lexical verb. Finally, the VP will front to Spec,CP, just as in (23), carrying the tense suffix and weak nominative enclitic along. This set of operations yields the input for the adjustment rule in (32). And since that set includes both narrowly syntactic and postsyntactic operations, the adjustment rule proposed here must apply very late in the derivation. This is in keeping with the general approach toward linearity-sensitive operations in the Distributed Morphology framework (see, for instance, Arregi and Nevins 2012): they must follow operations that refer to hierarchical syntactic structure.

6.3 Comparing the weak nominative enclitics against the strong pronouns

The weak nominative enclitics are partially homophonous with a separate set of strong pronouns; these strong pronouns are free from the rigid restrictions that circumscribe the distribution of the weak nominative enclitics. This partial homophony helps to explain why the two classes of morphemes were not differentiated in earlier descriptive work.

There is much morphosyntactic evidence that the strong pronouns, which are listed in Table 2, qualify as true nominals. For instance, they can bear the same range of case morphology attested on non-pronominal NPs. (33) illustrates with the instrumental case, which takes the shape -o after consonants and -m after vowels.

In (33a) the instrumental-marked nominal root is ha ‘this place, here’, yielding ham ‘to this place, hither’. In (33b) the instrumental case attaches to the strong pronoun ote ‘1pl.excl’. And in (33c) the instrumental case surfaces twice, once on the strong pronoun ẽrõ and once on the nominalizer hem. The latter morpheme subordinates the finite embedded clause ma’ã ’en herõwap hem ‘about the thing that you said yesterday’, which functions as an internally-headed relative. (See Section 4.2 for more examples of clausal nominalizations headed by hè$\sim$he.) That non-pronominal NPs and strong pronouns will bear the same case morphology in the same syntactic contexts strongly suggests that the strong pronouns are true nominals.

The strong pronouns can also take the plural-like collective suffix -’eat, widely attested on non-pronominal NPs: Tupari ‘Tuparí’ $\to$ Tupari’eat ‘the Tuparí people, the many Tuparí’; onõ ‘my friend(s), my relative(s)’ $\to$ onõ’eat ‘my many friends / relatives’. The ability of the strong pronouns to bear collective -’eat is shown by (34).

In contrast, speakers reject putting -’eat ‘many’ on the weak nominative enclitics.

Recall that the weak nominative enclitics never occur with present progressive, present habitual, or future auxiliaries (example 31, in Section 6.1). In contrast, the strong pronouns are allowed in any and all contexts.

(35) illustrates with second person forms. The strong pronouns in these three examples mark a change in topic and bear the nuclear case suffix.

The present and future auxiliaries in (35) are not compatible with the weak nominative enclitics, which require that tense be realized either as a second position particle or as a predicate-final suffix. Yet the nuclear-marked strong pronouns can and do occur here. This fact supports analyzing the strong pronouns as true arguments but weak nominative enclitics as agreement on T. Since the strong pronouns are not dependent on the T node the way that the weak nominative enclitics are, they may combine with the full range of tense marking.

Indeed, nuclear-marked strong pronouns can cooccur with weak nominative enclitics as long as the right kind of tense morphology is present. In (36a) and (36b), that morphology is the second position particle =ke ‘polite.future’.

The position of the weak nominative enclitic =’en ‘2sg’ is fixed: it must immediately follow the second position tense particle =ke ‘2sg’, which itself follows the clause-initial $wh$-word katkaere ‘when’. But the position of the strong pronoun ẽren ‘2sg + nuc’ is much more flexible. It may follow the second position cluster, as in (36a), or it may occur in a high, extrasentential position, as in (36b). In the latter case a prosodic break separates the extrasentential strong pronoun and the clause-initial $wh$-word, which sits in Spec,CP.

6.4 Summary and comparative discussion

The Tuparí language possesses two distinct sets of morphemes that were conflated in prior scholarship. Although the strong pronouns are partially homophonous with the weak nominative enclitics, the latter morphemes are parasitic on the second position particles and predicate-final suffixes, which they must linearly follow. They cannot, however, cooccur with the post-verbal auxiliaries that mark the same-day past, present and future. Whereas these auxiliaries paradigmatically contrast with the second position tense particles and the predicate-final tense suffixes, the weak nominative enclitics are highly restricted in their distribution. They are ‘just’ agreement, not true arguments.

Before turning to superficially tenseless clauses, in Section 7, we should address a marked difference between Tuparí and its best-described sister language, Sakurabiát (referred to in some prior literature as Mekens). According to Galucio (2014) (see also Galucio 2001, 38–43), subject pronouns are required in Sakurabiát only with transitive verbs. The following paradigm from Galucio (2014, 378; my highlighting) illustrates.

Sakurabiát does not require overt subject pronouns with intransitive verbs such as et ‘sleep’ but does require them with transitive verbs such as so(p) ‘see’. This pattern makes intuitive sense given the Tuparían languages' pattern of absolutive marking on lexical verbs. Intransitive verbs in the Tuparían languages always bear a pronominal prefix that conveys the person and number features of the subject. Hence the bolded pronouns in (37a) and (37b) are informationally redundant: they provide the same person-and-number information transmitted by the verb's pronominal prefix. However, as transitive verbs in the Tuparían languages take an object prefix, the bolded pronouns are required in (37c) and (37d): without those pronouns, there would be no information in these utterances about the person and number of the subject. For this reason the pronouns which are optional in (37a) and (37b) must be included in (37c) and (37d).

But Tuparí crucially differs from the Sakurabiát paradigm given in (37). In Tuparí, the presence of a weak nominative enclitic is conditioned entirely by the kind of tense marking employed, without regard to the valence of the lexical verb. Hence =’en ‘2sg’ is obligatory after the second position tense particle =ke ‘polite.future’ both in (38a), where the verb is intransitive ‘come.sg’, and in (38b), where the verb is transitive ‘give.’ Thanks to the absolutive argument marking pattern found on lexical verbs throughout the Tuparían family, the intransitive verb ‘come.sg’ in (38a) carries a pronominal prefix; and that prefix provides the same person-and-number information that the weak nominative enclitic does. Nonetheless, the weak nominative enclitic cannot be omitted.

As this minimal pair demonstrates, the valency of the lexical verb has no impact whatsoever on whether a weak nominative enclitic will be present. Rather, the distribution of the weak nominative enclitics is determined by the realization of T – a functional category which sits much higher than the VP in the clausal spine and which is therefore independent of the lexical verb's argument structure.

It is interesting to note that three subject pronouns – the first person singular, second person singular, and first person plural exclusive – may optionally encliticize onto the verb phrase in Sakurabiát (Galucio 2001, 41–43). This gives rise to alternations between SOV and OVS word order: the former order arises when the subject pronouns are clause-initial and the latter, when they have encliticized onto the VP. Similar variation between SOV and OVS is reported for another member of the Tuparían branch of the Tupían family, Wayoro (Nogueira 2011, 72–74; see also Nogueira 2019, ch. two). Tuparí appears to have undergone a more extreme version of the same encliticization process documented for Sakurabiát and Wayoro, with formerly independent pronouns now fully reanalyzed as Tense-parasitic agreement.

7.1 Null tense marking with verbal predicates

Verbal predicates that do not combine with overt tense marking receive one of two temporal interpretations, depending on their lexical semantics. A eventive/telic verb such as ko ‘eat, drink’ receives an immediately-prior-to-UT interpretation, while a stative/non-telic verb such as otetka ‘exceed, be taller than’ receives a present one.¹⁴ The near-minimal pair in (39) illustrates for the former case. In (39a), the clause-initial constituent is the verb phrase weut'eutkia and the instrumental-marked NPs arom and ’iporo occur after the weak nominative enclitic =’on. In (39b), it is the instrumental-marked NP which occurs clause-initially; the VP (bearing the adverbial focus suffix -ap) follows the weak nominative enclitic.

The weak nominative enclitic =’on ‘1sg’ surfaces in second position in both of these utterances; the syntactic category of the clause-initial constituent does not affect the enclitic’s placement. As examples like (39a) and (39b) receive a consistent temporal interpretation, they are justifiably analyzed as containing a null tense morpheme that contributes that interpretation. And as the weak nominative enclitic in these examples occurs in second position, that null tense morpheme must sit in second position as well. (40) provides revised glosses for (39).

The following near-minimal triple illustrates further. In each of these utterances the clause-initial constituent is a non-finite adverbial clause headed by -ro’are ‘while’, the weak nominative enclitic occurs in second position, and the main verb comes at the end of the clause and bears -ap/am ‘adv.foc’. Example (41c) must contain a null tense particle akin to overt =õpot ‘distant.past’ in (41a) and overt =ko ‘polite.future’ in (41b). This null particle ensures (a) that the weak nominative enclitic =’on ‘1sg’ will surface in second position and (b) that the utterance receives the correct temporal interpretation.

The appendix addresses an alternate placement pattern for weak nominative enclitics in verbal clauses with no overt tense marking.

7.2 Null tense marking with nominal predicates

Nominal predicates also cooccur with weak nominative enclitics (42) illustrates. In each of these utterances, a nominal predicate (soka ‘cold’, kisot'asap ‘our dying’, puop ‘knowledgeable, smart’) occurs with a weak nominative enclitic despite the absence of any overt tense morphology.

Given that the weak nominative enclitics are otherwise known to be dependent on a subset of tense marking, it must be the case that clauses like these contain more functional material than meets the eye. In other words, these clauses must contain a null T head.

Yet unlike the =$\varnothing$ ‘immediate.past’ proposed in Section 7.1, the null tense morpheme in (42a) through (43) cannot sit in second position. This is because the weak nominative enclitics in clauses with nominal predicates do not appear in second position; rather, they always follow the nominal predicate itself. This fact is clear from (42a) through (42c): in each of these three utterances a clause-typing particle follows the clause-initial constituent whereas the weak nominative enclitic follows the nominal predicate. Only when the nominal predicate is itself the clause-initial constituent will the weak nominative enclitic occur in second position. This is shown by (43), where puop ‘knowledgeable, smart’ is clause-initial. Examples such as this one are structurally parallel to (29), from Section 6, where the clause-typing particle linearly separates the predicate-final tense suffix from the weak nominative enclitic.

In sum, even though nominal predicates and superficially tenseless verbal predicates resemble one another in lacking overt tense morphology, they differ markedly in terms of the placement of the weak nominative enclitics.

To make sense of the placement of the weak nominative enclitics with nominal predicates we must take another look at the predicate-final suffixes -t/n ‘near.past’ and -pbi’a/mbi’a ‘durative’. Consider the following two pairs of examples, which build upon the generalizations discussed in Sections 6.1 and 6.2. In (44a) and (45a), the clause-typing particle =nẽ ‘yes/no’ follows the clause-initial constituent and the weak nominative enclitic =’en occurs immediately after the tense-marked VP. In (44b) and (45b), the tense-marked VP occurs as the clause-initial constituent and the weak nominative enclitic follows the clause-typing particle. That is, the enclitic =’en ‘2sg’ will cliticize directly onto the predicate as long as no clause-typing particle intervenes.

In terms of the placement of the weak nominative enclitics, clauses with nominal predicates (examples 42 and 43) behave just like clauses with verbal predicates where T is realized as a predicate-final suffix (examples 44 and 45). Given these parallels, the kind of null T head that is present in clauses with nominal predicates must be akin to the predicate-final suffixes -t/n ‘near.past’ and -pbi’a/mbi’a ‘durative’ rather than to the second position particles =ko/ke ‘polite.future’, =õpot ‘distant.past’, or =kut ‘ancient.past’. That is, this null T head must undergo postsyntactic Lowering to attach to the right edge of the predicate. This mechanism ensures that the weak nominative enclitic – independently known to depend parasitically on T – will follow the nominal predicate in the linear string.

(46) provides revised glosses for (42a) through (42b) and for (43), with the null tense suffix and the weak nominative enclitic now highlighted.

7.3 Enclitic placement in negated clauses confirms the existence of a null tense suffix on nominal predicates

The suffix -’om, which serves as both a negator (neg) and a privative (priv), maps nouns to nouns; negated constituents in Tuparí behave nominally rather than verbally according to a slew of morphosyntactic diagnostics (Singerman 2018a). If the generalizations given in the previous subsection about the placement of the weak nominative enclitics are accurate – and if negated predicates are just a kind of nominal predicate – then we predict that negated predicates should behave identically to all other nominal predicates as far as the nominative enclitics’ placement is concerned.

This prediction is correct. (47) illustrates. The weak nominative enclitics behave alike with negated predicates built from nominalized verbs (examples 47a and 47b) and with predicates where -’om attaches directly to a noun (47c and 47d).