Optimizing Display Other Formats:      PDF - 807k

Volume 7 Issue 1 (2009)        DOI:10.1349/PS1.1537-0852.A.333

Note: Linguistic Discovery uses Unicode characters to represent phonetic symbols. Please see Optimizing Display for requirements to accurately reproduce this page.

Covert Tense in Jarawara

Alan Vogel

SIL International

1. Introduction^[1]

The Jarawaras are an indigenous people group living within a reserve in the municipality of Lábrea in the state of Amazonas, Brazil. Their language, Jarawara, is a member of the small Arawá language family of southwestern Amazonia. Their small population of about 200 people requires that their language be classified as “endangered”, but their language has a high degree of vitality. Although the Jarawaras are fairly bilingual in Portuguese, the national language, they use their own language virtually exclusively for communication among themselves in their villages.

In this paper I argue that many Jarawara clauses that appear to have no tense morpheme may be analyzed as having a “covert” allomorph of what can be identified as immediate past eyewitness tense. The reason for the label “covert” is that, whereas immediate past eyewitness tense is typically realized as a suffix (as are all tense morphemes in Jarawara), in the clauses under consideration the exponent of immediate past eyewitness tense category is simply a particular type of gender agreement.

This idea is actually an extension of an analysis proposed by Dixon (2000, 2001, 2004) in his extensive work on Jarawara. The immediate past eyewitness suffix is -hara/-hare, ^[2] and the feminine variant of this suffix is used in (1).^[3]

Dixon showed that in a sentence such as (2), in contrast, the category of immediate past eyewitness tense is present without the suffix being present.

In contexts like those in (2), the suffix -hara/-hare is not allowed, yet it is quite clear that the category of immediate past eyewitness tense is present, not only from the meaning context in texts, but also from the paradigmatic relationship of verb forms like those in (1) and (2). The basic distribution rule is simple: If the “syntactic pivot” (in these sentences, the subject^[4] ) is first or second person plural, then the pattern in (2) is used instead of the suffix -hara/-hare.

What exactly “the pattern” is in sentences like (2) that indicates immediate past eyewitness tense, is a crucial question for this paper. Dixon (2000:27fn5) points to the repetition of the pronominal in the position after the verb stem. In (2) above, this is otaa in otaake. However, I argue in section 2 below that this repetition is required for any tense (actually, any tense-modal, see section 2 below), not just immediate past eyewitness tense, so it is not an indicator of this tense in particular. What indicates immediate past eyewitness tense in particular in (2) is the feminine gender agreement at the end of kama – if it were masculine, it would be kame .

Dixon applied the above analysis only to main clauses, such as (2) above. My purpose in this paper is to apply a similar analysis to a particular kind of subordinate clause, called a dependent clause (hence DC). The sentence in (3) consists of a main clause preceded by two DCs, and each of the three clauses is bracketed and labelled.

I will argue that the verbs of the two DCs in this sentence have immediate past eyewitness tense without having the -hara suffix, just as kama in (2) has immediate past eyewitness tense without the tense suffix, and that in both kinds of cases, the exponent of the tense category is a particular kind of gender agreement at the end of the verb, which is the vowel a in all three of these clauses, since there is feminine agreement.

One of the difficulties of this analysis is immediately apparent, since the verb in the main clause of this example (i.e. the final clause) has the far past eyewitness tense morpheme, -hamaro. How can the DCs be immediate past, if the main clause is far past? I will argue (in section 6 below) that the interpretation of tenses in subordinate clauses in Jarawara is different than in main clauses.

The paper is divided as follows. Sections 2 and 3 are introductory, providing general information about the tense-modal suffixes and about dependent clauses, respectively (the tenses are a subgroup of the tense-modals). Section 4 gives the full details of the gender agreement pattern referred to above. The most convincing evidence for my proposal is presented in section 5. There is a kind of agreement with possessors that is correlated with the presence of tense-modals, and this possessor agreement is found to be possible precisely in the DCs in which I am proposing there is covert immediate past eyewitness tense. Finally, section 6 presents ideas on how tense in DCs may be interpreted. While most DCs have what I am calling covert immediate past eyewitness tense, it is also true that many DCs have overt tense suffixes. It is obvious, though, that the tenses in DCs cannot be interpreted in the same way as those in main clauses, a fact recognized by Dixon as well. I propose generalizations that take into account covert immediate past eyewitness tense in DCs.

2. Jarawara Tense-Modals

2.1 General Information on Tense-Modals

Let us begin with the group of suffixes that Dixon calls “tense-modals”. These are listed in Table 1, which corresponds to Dixon’s (2004:197) Table 6.1.

Table 1. Forms of tense-modal suffixes

The first syllable in each case is in parentheses because it is often deleted by phonological rule. The last syllable of fut is in parentheses because the shortened form is used in certain morphological contexts.^[6] The reason for the two columns is to show that each suffix has a feminine and a masculine form. The suffixes are grouped in three rows according to semantic similarity. In the first row are the eyewitness past tenses, and in the second row the non-eyewitness past tenses. Furthermore, the first, second, and third items in each of these two rows correspond to each other, according to time frame. The third row is a somewhat more motley collection of tenses and modals that do not fit into the scheme of the first two rows.

The time frames of the different tenses are approximately as follows. Immediate past goes from the present back to up to a month or two ago. Recent past starts there and goes back to a year or two ago. Far past is anything before that.

Dixon calls the suffixes in table 1 the “tense-modal system”. There are dozens of other verbal suffixes in Jarawara, so what makes these a system, as opposed to other suffixes? It is clearly not the case that these suffixes occupy a single slot, since more than one can co-occur. In (4), for example -hemete FP.N co-occurs with -mone^[7] REP.

These suffixes do, however, occupy a slot in the sense that they come at a particular place in the predicate. Dixon calls this “slot G”, which is after the “miscellaneous suffixes” of slot F and before the “third pronominal position” of slot H. Furthermore, whenever there are two tense-modal suffixes, they are always adjacent, nothing else can come between them. These orderings are illustrated in the following two examples, each of which contains two suffixes from the tense-modals. In (5), -tee^[8] RP.N and -himona rep are preceded by -ma ‘back’, which is one of the “miscellaneous suffixes”. In (6), -hene irr and -mete FP.N are followed by the first person plural exclusive otaa, which is in the “third pronimal position”. (This is called the third pronominal position in contrast to the first and second positions, for object and subject agreement, respectively, which are to the left of the verb stem.)

Semantically, the past tenses (the first two rows in table 1) make a symmetrical system, with three time frames, and eyewitness and non-eyewitness variants for each time frame. The remaining suffixes in the third row don’t fit so nicely, but future, at least, can be considered a tense. The others have more modal meanings, and this is the reason for the label “tense-modals”. In fact, Dixon (2004:98) calls all five of these suffixes in the third row “modalities”.

Based on the instances of co-occurrence of tense-modals, I believe it is possible to propose a subdivision of the group into at least four slots. The proposed subdivision is in Table 2.

Table 2. Internal ordering of tense-modal suffixes.

While Dixon does not propose subdividing the tense-modals in this way, the orderings are based mostly on the types of co-occurrence which he (2004:196) describes: (1) FP.N and RP.N are very often followed by rep; (2) irr is attested followed by FP.N and by FP.e; and (3) fut can be followed by IP.N. In the following paragraphs, I consider each of these types of co-occurrence in turn, and then add a few others.

(1) Co-occurrence of FP.N and RP.N with REP is illustrated in (4) and (5) above, respectively.

(2) IRR is followed by FP.N in (6) above, and it is followed by FP.E in (7).

(3) FUT is followed by IP.N in (8).

In addition to these that are mentioned by Dixon, there are other similar co-occurrences. First, REP may follow INT, as in (9).

From this and from the cases in point (1) above, we may conclude that REP must follow some other tense-modals when it co-occurs with them.

Secondly, besides preceding FP.N and FP.E as mentioned in point (2) above ,IRR can also precede RP.E (10) and IP.N (11).

We can conclude that irr may precede a number of other tense-modals, and that it does not follow any others. This is why I have placed it in the first position in table 2.

Finally, in addition to following FUT, as mentioned above, IP.N can also follow REP, as in (12). For these reasons, it seems reasonable to put IP.N in a position by itself after all the other tense-modals.

IP.E does not co-occur with any other tense-modal, but since it is the eyewitness correspondent of IP.N, it seems reasonable to put it in the same position. This late position also seems consistent with Dixon’s (2001:27) idea that IP.E was innovated into the language later than the other tenses.

Summarizing, the evidence points to a division of tense-modals into at least four position classes, as in table 2. There are no ordering conflicts, such as would occur if one suffix could occur both preceding and following some other suffix. This internal organization, in turn, reinforces the idea that the tense-modals should be seen as filling a single (subdivided) slot, since no other suffixes may occur between the subdivisions.

2.2 Covert IP.E in Main Clauses

Dixon (2004:106) notes that in certain contexts, the specification of IP.E is accomplished without the use of the IP.E suffix -hara/-hare. (13), repeated from above, is such an example.

In a sentence such as this, it would be ungrammatical to have an overt IP.E suffix, i.e. *otaa kamahara otaake . Dixon explains that this phenomenon is limited to when the “syntactic pivot” (grammatical topic) is first or second person plural, the syntactic pivot being the subject of an intransitive or A-construction transitive, or the object of an O-construction transitive.^[9] (13) is intransitive, and the following examples illustrate the other two possibilities. The first clause of (14) is an A-construction, and (15) is an O-construction.

If there is no IP.E suffix, how do we know that sentences such as these have IP.E specified? Dixon points to the fact that the third pronominal position is occupied by the pivot argument. This is correct, in that sentences such as these lose their IP.E specification if the pronominal in the third position is removed. Compare, for example, the second clause of (16) with (17).

Note that while the translation of (15) above has a present tense in English, it is clearly IP.E in Jarawara. If this sentence were “tenseless” in Jarawara, it would be yara era mee wati kanake, i.e. without the pronominal in the third position. Jarawara IP.E sentences can often be translated as sentences with present tense in English. The quote in the following sentence (18) is another similar example from the same text. Even though the verb nafirarake has IP.E tense, the translation in English is present tense.

The presence of a first or second person plural pronominal in the third position is thus an indicator of IP.E tense. However, its importance as an indicator of this tense in particular can be overestimated, for two reasons. First, the third pronominal position is also filled for other tense-modal specifications besides IP.E, so it cannot be considered an exclusive indicator of IP.E tense. This is clear, for example in (19), repeated from above, cf. the ungrammatical *otaa kisamaroke.

The presence of a first or second person plural pronominal in third position is thus a marker of the presence of the tense-modal category in general, not of the IP.E choice specifically.^[11]

Secondly, there is a kind of gender agreement that distinguishes sentences with covert IP.E tense such as (13) from sentences that have no tense at all, overt or covert, such as (20).

Even though both kama in (13) above and obana in (20) end with a, it can be shown that the last a of kama shows feminine gender agreement, while the last a of obana does not, as Dixon (2004:106) notes. For verbs that end with a,^[12] the way to show this is to find a context in which the agreement is masculine, that is, where the vowel will be e instead of a. The suffix -ma in (13), for example, has the masculine form -me as in (21) to show masculine agreement.

It is impossible to have this kind of masculine agreement in a tenseless sentence like (20). That is, there are no main clauses in which a can become e for masculine agreement before a pronominal that is not first or second person plural.

For the verb stems that end with i or o, the situation is the reverse; that is, the contrast is in the sentences with feminine agreement. For feminine agreement a syllable ha is added for sentences that have covert IP.E tense, which is often reduced to just a (orthographic ya or wa, depending on whether the preceding vowel is i or o), cf. ohariya in (23) and tonafiyoha in (25). In contrast, there is no gender agreement before the pronominal in tenseless sentences, cf. ofimi (22) and osao (24).

In section 4 below I give a fuller description of this gender agreement pattern. But the preceding examples are sufficient to show that there is a kind of gender agreement that is characteristic of sentences with covert IP.E tense, which is not found in tenseless sentences.

3. Dependent Clauses

Dixon (2004) only discusses covert IP.E tense in relation to main clauses, and my proposal is that many subordinate clauses may also be analyzed in the same way. The kind of subordinate clause in question is called by Dixon a dependent clause (DC). A DC is almost always connected with an NP in the main clause of the sentence, and this is the case in (26), repeated again from above. In this sentence both DCs, which precede the main clause, have the same subject as the main clause.

There are two positions that DCs may occur in. The unmarked position is preceding the main verb, at the beginning of the sentence. This is the position of the two DCs in (26) above. The other position is following the main verb, at the end of the sentence. This is a marked position, as signalled by the pause that typically separates the DC from the main verb.^[14] Following Dixon, I will refer to these two positions as preposed and postposed. The final clauses of (27) and (28), repeated from above, are postposed DCs.

As these examples show, DCs are like main clauses in that they may or may not have overt tense-modals. The postposed DC in (27) has two tense-modals, but the postposed DC in (28) and the two preposed DCs in (26) have no overt tense-modal. I will argue that this similarity between DCs and tenseless main clauses is only apparent, and that the DCs that have no overt tense-modal in fact have a covert IP.E specification. Also unlike main clauses, DCs do not have mood morphemes such as the declarative marker -ke/-ka.

There are significant formal differences between preposed and postposed DCs, as is already suggested in the above three examples. First, the way the third pronominal position is filled is quite different. Note, for example, the presence of mati and ee at the ends of the postposed DCs in (27) and (28), respectively, but the absence of otaa at the ends of the two preposed DCs in (26). The only pronominal that can occur in the third position in a preposed DC is mee, as in this example from Dixon (2004:467).

In fact, Dixon (2004:466) argues that mee is required in the third position in a preposed DC whenever the pivot is this person (i.e. third person plural animate). This seems too stringent to me, since it would require that many subordinate clauses be classified as either relative clauses or juxtaposed clauses (see discussion below), so I prefer to view mee as being optional in third position in preposed DCs. However, the arguments about covert IP.E in this paper are unaffected by one’s point of view in this matter.

For postposed DCs, the other pronominals that occur besides mati and ee are owa (30), tiwa (31), otaa (32), and tee (33).^[15]

These pronominals do not occur in third position in preposed DCs. We have already seen in (26) that otaa does not occur, and similar examples can easily be produced for the non-occurrence of owa, tiwa, ee, tee, and mati as well.

Another difference between preposed and postposed DCs is that only postposed DCs can occur with the verbal suffix -haaro/-haari. The following example (34) contains tokens of both the masculine and feminine forms in successive clauses.

Dixon (2004:465) says that this suffix is added to a postposed DC if the pivot is third person singular, and if there is no tense-modal suffix.^[16] Another way to look at this, though, is to say that this suffix actually specifies IP.E tense. What suggests this is the fact that IP.E tense is otherwise conspicuously absent in postposed DCs (and in preposed DCs, too). But all the other tenses are found in postposed DCs. Dixon (2004:469, 470) notes that IP.N is common in postposed DCs, and that RP.E and FP.E are also found. Examples (35) and (36) show that RP.N and FP.N are also found in postposed DCs.

If -haaro/-haari is in fact IP.E, then we should expect some kind of correspondence with IP.N in postposed DCs. This is in fact the case, in that -haaro/-haari indicates eyewitness evidentiality, whereas IP.N indicates non-eyewitness evidentiality in these clauses. In the following paragraphs I consider all the examples of -haaro/-haari and all the examples of IP.N -hani/-hino in postposed DCs in the three texts in the appendix of Dixon’s (2004) grammar. Even though I have only selected the sentences that have -haaro/-haari or -hani/-hino in postposed DCs, the number of sentences is still large, so I have omitted the interlinear analysis, which in any case is provided in the original source, the appendix in Dixon’s grammar. I have, however, provided more literal translations in some cases, so that the clauses in the text can be paired with the corresponding clauses in the translation.

In Dixon’s first text there are two examples of -haari, and one example of IP.N -hino in a postposed DC. The story is about the death and burial of Siko, told by Manoware, who was present at the time. Below each sentence I provide a comment.

Dixon’s second text is about a Brazilian who while on an outing in the forest with the chief Okomobi, was stung by a bullet ant ( Paraponera clavata ) that got in his pants. All of the following postposed DCs except for two are instances of -haari, and all refer to events that Okomobi witnessed, so no comment on the individual sentences is necessary. The exceptions are two postposed clauses that have -hibanoho, the combination of future and IP.N, and I comment on these.

I assume that Okomobi did not actually see the ant stinging Salgado, but Okomobi’s use of eyewitness -haari in the above sentences is consistent with the sentence in which he talks about the ant stinging the Brazilian, which is also eyewitness (FPe):

I suppose the reason he uses eyewitness tense is that he saw the immediate effect of the ant stinging Salgado, i.e. his extreme discomfort.

In Dixon’s third text, Siko tells about being present when he was a boy when a bark canoe was made by his father and his companions. There are no occurrences of -haaro/-haari, and just one occurrence of IP.N -hani in a postposed DC. Some other Jarawaras came to see the canoe that had been made. Siko says they had been weaving fish traps, and he uses IP.N because he did not see them when they were weaving the fish traps.

The last clause in this sentence is also a postposed clause, and the verb has the intentive suffix -habone.

There is one asymmetry between -haaro/-haari and IP.N in postposed DCs that is suggested by these examples, and that is that whereas -haaro/-haari is only used for third person singular, IP.N may be used for any person. Note that IP.N is used with third person plural in the postposed DC in the last example. This is also true of other tenses (and other tense-modals besides the tenses). In the following example (50), RP.N is used in a postposed DC with the third person plural pronominal in third position.

This asymmetry is only apparent, however. I propose that the correspondent of -haaro/-haari for other persons is just gender agreement at the end of the verb stem, the same gender agreement we have already seen above in conjunction with covert IP.E tense in main clauses. In (51), owasiya shows feminine agreement, and in (52) repeated from above, ite show masculine agreement. As expected, these are eyewitness contexts.^[20]

In the next section I give a full description of this gender agreement.

Summarizing, my proposal for postposed DCs is that -haaro/-haari is the marker of IP.E in these clauses. It is used only to agree with a third person singular nominal.^[21] When agreement with any other person is demanded, there is covert IP.E, which is manifested the same way covert IP.E is manifested in main clauses, i.e. by the pronominal in third position and by gender agreement.

With this analysis in mind, we can now come back to the three texts in Dixon’s grammar, and look for additional postposed DCs that have IP.E, i.e. the ones in which the IP.E is covert. In the first text there is just one more postposed DC, and Manoware clearly witnessed the event.

There are two more postposed DCs in the second text. These also are eyewitness contexts.

In the third text there is just one more postposed DC, and this is an eyewitness context as well.

My proposal for analyzing preposed DCs is similar, except that with these there is no -haaro/-haari. There is only covert IP.E when this is the tense specification, for all persons. Before treating these, however, it should be observed that other tenses are possible in preposed DCs, and these are specified by the tense suffixes. Dixon (2004) gives a number of examples of IP.N tense in preposed DCs, including the following (57) in the first text in his appendix.

Other tenses are not at all common in preposed DCs,^[22] but here are examples with recent past non-eyewitness (58) and far past eyewitness (59) tenses, respectively.

By far the most common situation for preposed DCs is to have no overt tense-modal, and I am proposing that this is actually a covert specification of IP.E tense. Following are examples with various person specifications. (60) has a preposed DC with a third person plural pivot. The pivot is the subject, since it is an A-construction.

In (61) the pivot of the preposed DC is first person plural exclusive, since it is intransitive.

The pivot of the preposed DC in (62) is likewise the subject of an intrasitive, but it is third person singular masculine.

The preposed DC in (63) is an O-construction, so the object is the pivot. It is third person singular feminine.

These examples bear out the two differences between preposed and postposed DCs. First, the only pronominal that may occur in third position in a preposed DC is third person plural mee (60). In contrast, as we saw above a number of pronominals occur in third position in postposed DCs. Secondly, in clauses in which there would be -haaro/-haari in postposed DCs, this suffix does not occur in preposed DCs. This is true for (62) and (63). Putting these two facts together, we see that every one of these preposed DCs would have a different form if it were a postposed DC. The preposed DC in (60) would end with mati; the one in (61) would end with otaa; and those in (62) and (63) would end with -haari and -haaro, respectively.

There is, however, one thing that unifies nearly all these DCs, whether preposed or postposed, and that is gender agreement. If we exclude the DCs that have a tense-modal suffix (including -haaro/-haari, which I have proposed is a IP.E suffix), all the rest of the DCs have the same kind of gender agreement that we have seen is also characteristic of main clauses with covert IP.E tense. This is because they, too, have covert IP.E tense. In the next section I discuss more details about this gender agreement.

Table 3 summarizes my proposal as to how IP.E is indicated in main clauses, preposed DCs, and postposed DCs. There are three ways in which IP.E may be indicated, depending on what the person of the pivot is, and what kind of clause it is. They are (1) -hara/-hare, (2) gender agreement of the type I have described, and (3) -haaro/-haari.

Table 3. Marking of IP.E in main clauses and DCs.

For each context in the table I have also indicated whether there is a pronominal in third position, and if so, what the form of the pronominal is.

The comparison of the pattern in Table 3 with the pattern for the other past tenses (i.e. IP.N, RP.E and RP.N, and FP.E and FP.N) is interesting. For each of these other tenses, the pattern is quite simple: instead of there being three options for marking the tense, there is just one, i.e. the suffix for each tense in table 2 above (i.e. -hani/-hino for IP.N, and so on). And the pronominals in 3^rd position are exactly the same as those in table 3.

If this analysis is correct, then we must conclude that DCs all have a tense-modal specification, if not overt then covert. As we have seen above, this is not true for main clauses. Main clauses may be completely “tenseless”. This appears to be connected with the fact that main clauses are associated with mood morphemes such as declarative -ke/-ka, whereas in DCs mood morphemes are prohibited. That is, there apparently is a requirement that all finite clauses (i.e. main clauses and DCs)^[23] have either a tense-modal or a mood specification (and many main clauses have both).

I believe some cross-linguistic perspective can be gotten on DCs if Jarawara is seen as what Longacre (2007) calls a chaining language. Longacre divides the languages of the world into co-ranking languages on the one hand, and chaining languages on the other. This is how he describes this distinction (p. 375):

In co-ranking structures, such as those found in contemporary European languages, it is possible to have several verbs of the same rank, commonly referred to as independent verbs. Thus, we can speak of a sentence as consisting of a coordination of independent clauses. In English the conjunctions and, but, and or, plus a few others, join such independent clauses into sentence units…. In a chaining structure, on the other hand, it is simply not possible to join two such verbs of the same rank in the same sentence. A sentence either ends in a dominating verb of fuller structure than that of the preceding verbs, or alternatively, begins with a dominating verb of fuller structure than that of the following verbs. In the former case, the preceding verbs of restricted structure are often referred to as medial verbs (or as participles, gerunds, or even coverbs) while the dominating verb at the end is referred to as the final verb. In the latter case, the following verbs of restricted structure are referred to as consecutive (or sequential) verbs while the dominating verb at the beginning is referred to as the initial verb. In the former case we speak of medial-final chaining; in the latter case we speak of initial-consecutive chaining.

According to this view, Jarawara would be medial-final chaining language.

Longacre gives a number of characteristics of this kind of language, and Jarawara fits the profile quite well. First, it should be clear from the data presented so far that main clauses in Jarawara are quite distinct from the preposed DCs that precede them, in that they typically have a mood morpheme, whereas the preposed DCs cannot have a mood morpheme. Furthermore, it is common to have not just one but a whole series of preposed DCs preceding the main clause. In (141) below, for example, there are four preposed DCs in a row, and this sort of thing is not uncommon in Jarawara texts.^[24] Longacre says “the final clause is like an engine that pulls a string of cars,” and this is a good description of many Jarawara sentences.

The existence of postposed DCs is not a problem for this analysis. As stated above, postposed DCs are clearly in a marked position, as opposed to preposed clauses, which are in an unmarked position. For this reason, Longacre’s label “medial clause” would probably be a more accurate way of talking about the preposed DCs of Jarawara, since preposed suggests out of place.

A second characteristic of medial-final chaining languages that Longacre gives is that they are OV/head-final languages, in contrast to initial-consecutive chaining languages, which have the basic sentence constituent order VO. In Vogel (2003:80f) I discuss several kinds of evidence that indicate that Jarawara is OV and head-final.

Jarawara appears to be exceptional among medial-final chaining languages in one respect, and that is that Jarawara does not have a switch-reference system. Longacre (p. 399) says in medial-final chaining languages, the medial clauses are marked to indicate whether the following clause (or in some cases, the final clause) has a different subject. It is true that the A-construction/O-construction contrast in Jarawara performs a similar function as switch-reference systems of other languages, but only in the very general sense of helping to track participants. But the differences seem to be more significant than the similarities. For one thing, the A-constructions and O-constructions in Jarawara are used to track discourse topics, not subjects. Also, the A-construction/O-construction contrast only applies to transitive clauses; intransitives can only be one way, since the pivot can only be the subject. In contrast, in a switch-reference system, an intransitive can be marked as either same-subject or different subject.

What is the value of Longacre’s theory for Jarawara, in the present context? One of the ways in which the theory is born out in Jarawara is that, not only do preposed DCs not have mood, in contrast to main clauses. They also are marked for tense in very different ways than main clauses, and one of the ways is that marking for tense is reduced, as the theory predicts. If my proposal is correct, then most DCs have IP.E tense, whereas they occur with main clauses that often have different tense specifications. In section 6 below I return to the issue of tense marking and the interpretation of tense in DCs.

There are at least two other kinds of subordinate clauses that have the same kind of agreement in the verb stem as DCs. One of these is relative clauses, and the other is juxtaposed clauses, both of which have been described by Dixon (2004).

Relative clauses in Jarawara are formally the same as preposed DCs, except for not having a third pronominal position.^[25] For this reason it is not easy to distinguish relative clauses from preposed DCs, and in fact it may be possible to analyze most preposed DCs as relative clauses.^[26] However, there are certain clauses that must be analyzed as relative clauses, because they are clearly part of an NP. The kind of context in which this is most readily apparent is in conjunction with certain morphemes that may only be attached to NPs. Usually they are attached to nouns, but since they are attached to whatever the last word of the NP is, they can be attached to a verb if it is part of a relative clause. One of these morphemes is -ra, which is an object marker. In (64), -ra is attached to a possessed noun, and in (65) it is attached to an adjective. In both cases, it occurs at the end of the NP which is the object of the clause.

In (66), however, -ra occurs at the end of a verb. The clause fara kawiyabanira ‘the starch that was eaten with herself’ is the object of the sentence; it is a relative clause.

The verb in (66) has tense-modals, but the verb that -ra is attached to in (67) has no overt tense-modal. According to my analysis, it has IP.E tense, as shown by the masculine agreement.

Other morphemes that are only attached to NPs are taa ‘contrast’ (68), nima ‘like’ (69), and ni yaa ‘to’ (70). In these sentences these morphemes are clearly attached to verbs, but it is because the verbs are part of relative clauses.

There are other contexts in which these kinds of clauses are clearly part of NPs and thus must be analyzed as relative clauses, and Dixon (2004:525f) includes additional discussion; but these will suffice to demonstrate the phenomenon. The point I want to make is that relative clauses, when they do not have an overt tense-modal, have the same gender agreement pattern as preposed DCs, and the analysis of covert IP.E in DCs applies to them as well. Just as DCs may be analyzed as having the tense-modal category, all relative clauses may be analyzed this way as well.

Juxtaposed clauses encode various related semantic relations such as frustration and counter expectation, which usually may be translated using the conjunction “but” in English. The structure is a subordinate clause followed by a main clause. Typically the subordinate clause has the intentive tense-modal -habone/-hibona (71), but it is also common for there to be no overt tense-modal (72) (73). In this case the agreement pattern characteristic of covert IP.E occurs.

Dixon (2004:529) says that the meaning “and” may also be encoded by juxtaposition. The example he gives is (74).

This may be an unnecessary broadening of the scope of juxtaposition in Jarawara. I analyze this sentence as just a preposed DC followed by a main clause.^[27] Dixon cannot analyze these sentences this way because, as I discuss in the next section, he only accepts as preposed DCs clauses that have a -ha/-hi suffix (or mee, if the pivot is that person). In my view, the -ha/-hi marker is optional, so there is no problem in analyzing the first clause of (74) as a preposed DC. There is, however, one structural difference between juxtaposed clauses and preposed DCs, and that is that juxtaposed clauses (like relative clauses, cf. above) cannot have mee in the third pronominal position, whereas preposed DCs can. That is, juxtaposed clauses do not have a third pronominal position at all.

Whatever definition of juxtaposed clause is accepted, the point for the present analysis is the same: like relative clauses, juxtaposed clauses that do not have an overt tense-modal have the same gender agreement pattern as DCs, and they therefore may be seen as having covert IP.E specified.

4. Gender Agreement in Tensed Clauses

I have proposed that there is a certain kind of gender agreement that is characteristic of covert IP.E tense contexts. These contexts include all DCs that have no overt tense-modal (recall that I have defined -haaro/-haari as a tense-modal); and main clauses that have no overt tense-modal, and that have a first or second person plural pronominal in the third position. The pronominals are 1IN ee, 1EX otaa, and 2PL tee. Unlike DCs, main clauses may occur completely without the tense-modal category, in which case these pronominals do not occur in third position.

The gender agreement in question occurs at the end of the verb stem. The verb stem for this purpose is as follows. For inflecting verbs that consist of a bare root, it is the root. For non-inflecting verbs, i.e. verbs that require an auxiliary na or ha, it includes the auxiliary. For verbs (both inflecting and non-inflecting) that have miscellaneous suffixes, it includes the miscellaneous suffixes. The verb stem for this purpose also includes the negative suffix -ra, which is not a miscellaneous suffix but follows miscellaneous suffixes.

Phonologically speaking, there are not very many ways a word can end in Jarawara, since there are no closed syllables, and there are only four vowels, a, i, e, and o. There are three complicating factors, however, all of them connected in one way or another to the phoneme e. First, there is a morphophoneme I that is realized on the surface as either i or e, depending on whether the preceding number of moras in the word is odd or even. Secondly, all inflecting verb roots that end with e (the phoneme, not the e realization of I) have a syllable ha added to the underlying root. This ha is deleted if the preceding number of moras in the word is odd, but it is maintained if the preceding number of moras is even. Thirdly, the habitual suffix -tee has a long vowel, and when it comes at the end of the phonological word there is usually no gender agreement, although -ha/-hi is occasionally added.

By far the most common vowel for verb stems to end with is a. The form for feminine agreement is a, and the form for masculine agreement is e. We have seen many examples of this alternation above, so I will not repeat them here.

There is, however, a slight variation on the a/e theme, when in some cases an actual marker is added, of the form -ha/-hi.^[28] The feminine form -ha is contained in famaha in (75), repeated from above.

The masculine form, -hi, is illustrated in (76), in the word awahi .

When -hi is used, normally a preceding a does not change to e, so that the sequence is ahi as in (76) above. But occasionally both e and -hi are used together, giving the sequence ehi, as in the word yanehi in (77).

Dixon (2004:466) calls -ha/-hi the marker of preposed DCs that do not have mee in third pronominal position, but the fact is that this syllable can optionally occur in many other contexts as well. For one thing, it can occur in a preposed DC that has mee in third position (78), whereas Dixon does not allow for this possibility.

The -ha/-hi ending can also occur in postposed DCs, as we have seen in (75) above; and it can occur in relative clauses as well, as in (79) and (80).^[29]

The -ha/-hi ending may even occur in main clauses, as in (81).

In short, the -ha/-hi ending is basically just a variation of the a/e gender agreement pattern. And since it is optional in each of the contexts in which it occurs, it should not be seen as a requirement for preposed DCs.^[30]

For stems ending in i, feminine agreement is shown by iha or the variant ia, which is orthographically iya.^[31] These are shown, for example, in the words kasawar iha and owas iya in (82) and (83), respectively, repeated from above. Masculine agreement is shown by ihi or just i. These are illustrated in fimihi and ohari in (84) and (85) below, respectively.

When the morphophoneme I is realized as i, the agreement pattern is the same as that for i. Note the alternatives for feminine agreement iha in (86) and iya in (87), and i for masculine agreement in (88). There happens to be no ihi for masculine agreement in my data, but I expect this variant to be revealed in further data.

When the morphophoneme I is realized as e, -ha can be added for feminine agreement (89) and -hi for masculine agreement (90); but it is normal for these not to be added, and in this case there is no difference between feminine and masculine agreement. Kake in (91) has a masculine subject, and kakisake in (92) has a feminine subject, but they both end in e, not ehi or eha.

As stated above, it can be argued that there are no verb roots ending in the phoneme e, since there is an underlying ha at the end of these roots. The ha is present if the preceding number of moras in the word is even, and it is deleted if the preceding number of moras is odd. This being the case, roots ending in e do not completely parallel those ending with the e realization of I the same way those ending in i do parallel those ending with the i realization of I.

That the ha syllable is an underlying part of the root of these verbs can be shown by a comparison of examples such as the following two. In nakomeka in (93), the preceding number of moras is odd (i.e. three), and so the ha is deleted; but in onakome hateere in (94), the preceding number of moras is even (i.e. four), so the ha is retained.

In covert IP.E contexts, where there is feminine agreement, the ha is always retained, even when the number of preceding moras in the word is odd. Compare for example keye ha in (95) and okeyeha in (96).

For contexts in which there is masculine agreement, the ha can become he, as in hinakome he in (97); and if the number of preceding moras is odd, the ha is deleted, and a - hi may be added for agreement, as in hikeyehi in (98).

As mentioned above, there is one verbal suffix that ends with a long e, i.e. the habitual suffix -tee. This often occurs at the end of the phonological word, but it is usually not associated with the type of gender agreement I have described. This seems to be just a phonological characteristic of -tee because of the long e; but it is not an indication that covert IP.E is not associated with -tee. The following examples are typical. In (99), -tee comes at the end of a relative clause, and the head is masculine, but there is no masculine agreement. There is likewise no feminine agreement in (100), even though the postposed DC that -tee is part of has a feminine pivot (because it is animate plural). Thus there is no gender agreement whether the pivot is feminine or masculine.

Occasionally, though, there are examples such as the following pair, in which-tee is followed by ha for feminine agreement (101), and hi for masculine agreement (102).

A fact that needs to be kept in mind in relation to -tee is that there are contexts in which it is found with overt IP.E, so it is clearly not incompatible with it. It is found both with the main clause suffix -hara/-hare (103) and with the -haaro/-haari suffix that is found in postposed DCs, which I have characterized as an IP.E marker (104).

The last vowel to consider is o. For stems that end in o, in DCs ha is added for feminine agreement, as in kamoha in (87) above. This may be pronounced as just a, and is then written wa in the orthography, as in hikiyowa in (105).

For agreement with a masculine nominal, -hi can be optionally added, as in watohi in (106). However, the -hi is not required, as shown by hikiyo in (107).

This completes the description of the gender agreement pattern at the end of the verb stem when there is covert IP.E. Before moving on, though, it needs to be observed that there is one other context in which this same gender agreement occurs that I have not mentioned yet, and that is in main clauses with what Dixon (2004) has called a “secondary verb”.

There are two secondary verbs, ama and awine/awa. They occur in “Slot I” of the predicate, immediately following the pronominal position discussed above.^[33] They are not suffixes, but phonologically independent words. We have already seen numerous examples of these morphemes above, and I repeat several of them below: ama in (108), awine in (109) and awa in (110).

As these examples show, awine/awa is a kind of evidential, with a meaning something like “it appears that” or “it seems”. The meaning of ama is harder to pin down, as Dixon (2004:228) also notes. He says it generally has the meaning “extended in time,” and this is compatible with (108). I would add that when used without any tense-modal, it more specifically describes a situation that was witnessed by the speaker, and usually a situation that obtained in the recent past but no longer obtains. This use is illustrated in (111) and the second sentence of (112).

Ama is also used quite frequently in conjunction with certain other morphemes, and most of these cases are compatible with the “extended in time” meaning. For example, it is rare for the habitual suffix -tee to occur in a main clause verb without ama, as in (113), repeated from above.

Ama is also practically obligatory with irr in a main clause, but here the “extended in time” meaning may not be very relevant. In the above examples, this meaning seems to be relevant in (108) above, but not in (114), (115), or (116), all repeated from above.